{"version":"6.14.0","api_prefix":"/gradio_api","mode":"blocks","app_id":5836739678550159256,"dev_mode":false,"vibe_mode":false,"analytics_enabled":true,"components":[{"id":1,"type":"html","props":{"_retryable":false,"_undoable":false,"likeable":false,"streamable":false,"value":"
Hume AIHume AI

Real World VoiceEQ Benchmark

","html_template":"${value}","css_template":"","js_on_load":"element.addEventListener('click', function() { trigger('click') });","apply_default_css":true,"show_label":false,"visible":true,"elem_classes":[],"preserved_by_key":["value"],"container":false,"padding":false,"autoscroll":false,"buttons":[],"props":{},"name":"html","_selectable":false,"component_class_name":"HTML"},"skip_api":false,"component_class_id":"42fa1133b1928d16a86f4bd1a5ac9b444ba8baf37b247373f67eca5a7420b7ee","key":null,"api_info":{"type":"string"},"api_info_as_input":{"type":"string"},"api_info_as_output":{"type":"string"},"example_inputs":"

Hello

"},{"id":2,"type":"tabs","props":{"visible":true,"elem_classes":[],"preserved_by_key":[],"name":"tabs"},"skip_api":true,"component_class_id":"09c75f0eb61f20e7fc9c769c6f7a6fa5c820f355ffffdf6b7e4f1f26e737d60c","key":null},{"id":3,"type":"tabitem","props":{"label":"About","visible":true,"interactive":true,"elem_classes":[],"preserved_by_key":[],"render_children":false,"name":"tab"},"skip_api":true,"component_class_id":"8ceb480a50807334325f24b3a91000911bf4db795c2f0afa24fc033d7574a964","key":null},{"id":4,"type":"html","props":{"_retryable":false,"_undoable":false,"likeable":false,"streamable":false,"value":"

The Real World VoiceEQ Benchmark evaluates more than 40 leading proprietary and open-source voice models across 15+ evaluation areas and 60+ metrics spanning Automatic Speech Recognition (ASR), Text-to-Speech (TTS), Speech-to-Speech (S2S), and Speech Understanding. Built on 700,000+ human evaluations across diverse accents, ages, speaking styles, emotions, and acoustic environments, it is one of the largest human evaluations of voice AI conducted to date. Every evaluation was powered by Kairos, our voice-native evaluation platform.

The benchmark is organized into separate leaderboard panels — click one to open it:

Text-to-Speech (TTS)→

Measures how naturally, expressively, and accurately models convert text into speech across real-world scenarios.

Speech-to-Speech (STS)→

Measures how well voice agents understand spoken language and generate helpful, emotionally appropriate spoken responses.

Speech Understanding→

Measures how well models understand meaning beyond the words, including emotion, intent, speaker characteristics, and conversational context.

Automatic Speech Recognition (ASR)→

Measures transcription accuracy across diverse speakers, accents, emotions, background noise, and conversational conditions using real-world audio and Word Error Rate (WER).

Each leaderboard contains targeted evaluation categories designed to surface failures that are often missed by short, clean, neutral examples.

For the technical report, please click here.

","html_template":"${value}","css_template":"","js_on_load":"element.addEventListener('click', function() { trigger('click') });","apply_default_css":true,"show_label":false,"visible":true,"elem_classes":[],"preserved_by_key":["value"],"container":false,"padding":false,"autoscroll":false,"buttons":[],"props":{},"name":"html","_selectable":false,"component_class_name":"HTML"},"skip_api":false,"component_class_id":"42fa1133b1928d16a86f4bd1a5ac9b444ba8baf37b247373f67eca5a7420b7ee","key":null,"api_info":{"type":"string"},"api_info_as_input":{"type":"string"},"api_info_as_output":{"type":"string"},"example_inputs":"

Hello

"},{"id":5,"type":"tabitem","props":{"label":"Text-to-Speech","visible":true,"interactive":true,"elem_classes":[],"preserved_by_key":[],"render_children":false,"name":"tab"},"skip_api":true,"component_class_id":"8ceb480a50807334325f24b3a91000911bf4db795c2f0afa24fc033d7574a964","key":null},{"id":6,"type":"markdown","props":{"value":"Evaluates whether models can turn text into speech that sounds expressive, believable, accurate, and stable across varied real-world situations.\n\nIncludes audiobook narration, sports commentary, stand-up comedy, delivery of varied emotions, code-switching, heteronyms disambiguation, pharmaceutical terms, alphanumeric strings, structured text, long-form reading, and overall audio fidelity.","show_label":false,"rtl":false,"latex_delimiters":[{"left":"$$","right":"$$","display":true}],"visible":true,"elem_classes":[],"preserved_by_key":["value"],"sanitize_html":true,"line_breaks":false,"header_links":false,"container":false,"padding":false,"name":"markdown","_selectable":false},"skip_api":false,"component_class_id":"c2f35e73b9c0ffb3ea6b3682c71039be03350e8cd5cf185fdbbc33d52e7a2945","key":null,"api_info":{"type":"string"},"api_info_as_input":{"type":"string"},"api_info_as_output":{"type":"string"},"example_inputs":"# Hello!"},{"id":7,"type":"row","props":{"variant":"default","visible":true,"elem_classes":[],"equal_height":false,"show_progress":false,"preserved_by_key":[],"name":"row"},"skip_api":true,"component_class_id":"bc47bf2e53002b34f9ec79dc4a3052e03866c51adfd499cc985f9fa94a4de216","key":null},{"id":8,"type":"radio","props":{"choices":[["all","all"],["open-source","open-source"],["proprietary","proprietary"]],"value":"all","type":"value","label":"License","show_label":true,"container":true,"min_width":160,"visible":true,"elem_classes":[],"preserved_by_key":["value"],"rtl":false,"buttons":[],"name":"radio","_selectable":false},"skip_api":false,"component_class_id":"112db8570d794fa26decdf4f78c1582d4b272e0f664857db79971379947254a0","key":null,"api_info":{"enum":["all","open-source","proprietary"],"title":"Radio","type":"string"},"api_info_as_input":{"enum":["all","open-source","proprietary"],"title":"Radio","type":"string"},"api_info_as_output":{"enum":["all","open-source","proprietary"],"title":"Radio","type":"string"},"example_inputs":"all"},{"id":9,"type":"form","props":{"scale":1,"min_width":160,"preserved_by_key":[],"name":"form"},"skip_api":true,"component_class_id":"7c96e70477dd55732fb1a4bc6696ddcc8ff96faeabe823f8c3311f0d1f92d0eb","key":null},{"id":10,"type":"html","props":{"_retryable":false,"_undoable":false,"likeable":false,"streamable":false,"value":"
15mean human rating
#providerLicenseActing / Role-fit ▼ⓘExpressiveness ↕ⓘVoice identity ↕ⓘLanguage Stability ↕ⓘReliability ↕ⓘExtended Long-Form Speaker Stability ↕ⓘLong-form Speaker Stability ↕ⓘAcoustic Quality ↕ⓘ
1google/gemini-3.1-flashproprietary3.973.643.003.273.821.223.273.15
2google/gemini-2.5-flashproprietary3.643.442.593.253.662.193.393.28
3google/gemini-2.5-proproprietary3.633.322.803.283.892.903.583.27
4openai/tts-1*proprietary3.163.213.463.202.47††3.303.33
5qwen/Qwen3-TTS-12Hz-1.7Bopen-source3.153.082.672.522.823.323.023.19
6microsoft/VibeVoice-1.5B*open-source3.142.943.112.682.873.193.072.75
7openai/tts-1-hd*proprietary3.113.163.463.132.63††3.323.23
8sesame/csm-1b*open-source3.093.012.042.732.813.443.062.61
9smallest/lightning_v3.1proprietary3.083.303.062.823.113.253.113.15
10nari-labs/diaopen-source3.082.122.642.602.303.012.281.22
11elevenlabs/eleven_v3*proprietary3.063.172.863.333.593.083.083.26
12fishaudio/s2-pro*open-source3.042.933.533.482.883.503.303.34
13hume/tada-3b-mlopen-source2.933.062.673.422.21·2.482.83
14deepgram/aura-2*proprietary2.923.223.162.203.60††2.913.30
15cartesia/sonic-3.5*proprietary2.913.143.532.453.493.533.143.37
16openai/gpt-4o-mini-ttsproprietary2.883.233.363.343.493.003.133.34
17elevenlabs/eleven_multilingual_v2*proprietary2.873.113.402.912.633.332.933.29
18bosonai/higgs-audio-v3-tts-4bopen-source2.832.012.223.383.202.532.842.71
19bosonai/higgs-audio-v2open-source2.793.151.702.962.753.272.813.18
20resemble-ai/chatterboxopen-source2.772.782.923.242.72††2.912.78
21inworld/tts-1-max*proprietary2.763.053.273.553.16††2.863.32
22hexgrad/Kokoro-82M*open-source2.732.963.502.012.533.543.033.28
23inworld/tts-1*proprietary2.702.893.153.223.24††2.733.31
24coqui/xtts-v2*open-source2.682.813.383.112.55††2.802.86
25inworld/tts-2*proprietary2.592.893.252.943.49††2.943.33
26parler-tts/parler-ttsopen-source2.572.611.802.812.092.162.082.57
27indexteam/indextts-2*open-source2.511.833.422.973.013.523.301.01
* Some evals across the leaderboards include questions related to instruction-tuning which this model was not explicitly trained to do
†† Generation terminated after reaching the maximum token limit.
","html_template":"${value}","css_template":"","js_on_load":"element.addEventListener('click', function() { trigger('click') });","apply_default_css":true,"show_label":false,"visible":true,"elem_classes":[],"preserved_by_key":["value"],"container":false,"padding":false,"autoscroll":false,"buttons":[],"props":{},"name":"html","_selectable":false,"component_class_name":"HTML"},"skip_api":false,"component_class_id":"42fa1133b1928d16a86f4bd1a5ac9b444ba8baf37b247373f67eca5a7420b7ee","key":null,"api_info":{"type":"string"},"api_info_as_input":{"type":"string"},"api_info_as_output":{"type":"string"},"example_inputs":"

Hello

"},{"id":11,"type":"html","props":{"_retryable":false,"_undoable":false,"likeable":false,"streamable":false,"value":"

Sample Generations

Humorous Delivery
My boss told me to have a good day, so I went home.
Better
Worse
Sports Commentary
No, we got a free safety blitz off the edge. He changed the play. The give is to Lynch. Up the middle and push back short. If you're running the ball on third and three, that's telling me you're going for it on fourth down. It's Lynch. They got him by the ankles for a moment. He fights for it. And if he comes downhill, they're going to throw this fade down at the bottom.
Better
Worse
","html_template":"${value}","css_template":"","js_on_load":"element.addEventListener('click', function() { trigger('click') });","apply_default_css":true,"show_label":false,"visible":true,"elem_classes":[],"preserved_by_key":["value"],"container":false,"padding":false,"autoscroll":false,"buttons":[],"props":{},"name":"html","_selectable":false,"component_class_name":"HTML"},"skip_api":false,"component_class_id":"42fa1133b1928d16a86f4bd1a5ac9b444ba8baf37b247373f67eca5a7420b7ee","key":null,"api_info":{"type":"string"},"api_info_as_input":{"type":"string"},"api_info_as_output":{"type":"string"},"example_inputs":"

Hello

"},{"id":12,"type":"tabitem","props":{"label":"Speech-to-Speech","visible":true,"interactive":true,"elem_classes":[],"preserved_by_key":[],"render_children":false,"name":"tab"},"skip_api":true,"component_class_id":"8ceb480a50807334325f24b3a91000911bf4db795c2f0afa24fc033d7574a964","key":null},{"id":13,"type":"markdown","props":{"value":"Evaluates whether voice agents can understand spoken input and respond with speech that is useful, emotionally attuned, calm under pressure, natural, and stable across varied real-world interactions.\n\nIncludes emotionally ambiguous confirmations, expression-gain comparisons between audio and transcript, degraded phone-line tasks, urgent callers, hostile callers, and scenarios where the agent must remain helpful while managing conversational stress.","show_label":false,"rtl":false,"latex_delimiters":[{"left":"$$","right":"$$","display":true}],"visible":true,"elem_classes":[],"preserved_by_key":["value"],"sanitize_html":true,"line_breaks":false,"header_links":false,"container":false,"padding":false,"name":"markdown","_selectable":false},"skip_api":false,"component_class_id":"c2f35e73b9c0ffb3ea6b3682c71039be03350e8cd5cf185fdbbc33d52e7a2945","key":null,"api_info":{"type":"string"},"api_info_as_input":{"type":"string"},"api_info_as_output":{"type":"string"},"example_inputs":"# Hello!"},{"id":14,"type":"row","props":{"variant":"default","visible":true,"elem_classes":[],"equal_height":false,"show_progress":false,"preserved_by_key":[],"name":"row"},"skip_api":true,"component_class_id":"bc47bf2e53002b34f9ec79dc4a3052e03866c51adfd499cc985f9fa94a4de216","key":null},{"id":15,"type":"radio","props":{"choices":[["all","all"],["open-source","open-source"],["proprietary","proprietary"]],"value":"all","type":"value","label":"License","show_label":true,"container":true,"min_width":160,"visible":true,"elem_classes":[],"preserved_by_key":["value"],"rtl":false,"buttons":[],"name":"radio","_selectable":false},"skip_api":false,"component_class_id":"112db8570d794fa26decdf4f78c1582d4b272e0f664857db79971379947254a0","key":null,"api_info":{"enum":["all","open-source","proprietary"],"title":"Radio","type":"string"},"api_info_as_input":{"enum":["all","open-source","proprietary"],"title":"Radio","type":"string"},"api_info_as_output":{"enum":["all","open-source","proprietary"],"title":"Radio","type":"string"},"example_inputs":"all"},{"id":16,"type":"form","props":{"scale":1,"min_width":160,"preserved_by_key":[],"name":"form"},"skip_api":true,"component_class_id":"7c96e70477dd55732fb1a4bc6696ddcc8ff96faeabe823f8c3311f0d1f92d0eb","key":null},{"id":17,"type":"html","props":{"_retryable":false,"_undoable":false,"likeable":false,"streamable":false,"value":"
15mean human rating
#providerLicenseEmotion Understanding ▼ⓘEmotion Alignment ↕ⓘExpressive Robustness ↕ⓘVoice naturalness ↕ⓘProblem redirecting ↕ⓘ
1openai/gpt-live-1proprietary3.373.643.883.913.48
2inworld/cascade-3‡‡proprietary3.363.023.382.523.39
3openai/gpt-realtime-2proprietary3.343.743.802.803.47
4deepgram/cascade-1††proprietary3.283.143.543.943.66
5google/gemini-2.5-flash-native-latestproprietary3.263.674.024.12‖‖
6google/gemini-3.1-flash-liveproprietary3.243.763.894.193.77
7deepgram/cascade-2*proprietary3.202.903.584.273.66
8openai/gpt-realtime-miniproprietary3.173.143.663.263.44
9inworld/cascade-4§§proprietary3.133.013.322.323.57
10qwen/qwen3-omni-30b-a3bopen-source3.133.192.252.062.91
11xai/grok-voice-latestproprietary3.073.413.803.133.61
12elevenlabs/agentproprietary2.973.452.892.603.40
13nvidia/personaplexopen-source2.631.842.582.611.71
14moonshotai/kimi-audio-7b-instructopen-source2.322.171.672.752.58
15kyutai/moshikoopen-source1.361.001.001.031.00
* deepgram/cascade-2 = nova-3 → gpt-4o → aura-2-thalia-en
†† deepgram/cascade-1 = nova-3 → gpt-4o-mini → aura-2-thalia-en
‡‡ inworld/cascade-3 = inworld-stt-1 → gpt-4o → inworld-tts-2
§§ inworld/cascade-4 = inworld-stt-1 → gpt-4o-mini → inworld-tts-2
‖‖ The model’s guardrails rejected the interaction and did not generate a response to the user.
","html_template":"${value}","css_template":"","js_on_load":"element.addEventListener('click', function() { trigger('click') });","apply_default_css":true,"show_label":false,"visible":true,"elem_classes":[],"preserved_by_key":["value"],"container":false,"padding":false,"autoscroll":false,"buttons":[],"props":{},"name":"html","_selectable":false,"component_class_name":"HTML"},"skip_api":false,"component_class_id":"42fa1133b1928d16a86f4bd1a5ac9b444ba8baf37b247373f67eca5a7420b7ee","key":null,"api_info":{"type":"string"},"api_info_as_input":{"type":"string"},"api_info_as_output":{"type":"string"},"example_inputs":"

Hello

"},{"id":18,"type":"html","props":{"_retryable":false,"_undoable":false,"likeable":false,"streamable":false,"value":"

Sample Conversations

Scenario: Bank-transfer confirmation, hesitant stance. The caller's authorization is a single ambiguous “yeah…” — real recorded audio, deliberately unclear. After the call each agent is pulled out of character and asked what tone it heard. Rated by 12 humans on a 1–7 scale across two questions: initial response appropriate (turns 1–5) and interpretation accurate (turn 7). Turn 4 is the identical clip in both calls.
Poorly rated conversation3.63 / 7 human rating
Initial response appropriate 4.75 · Interpretation accurate 2.50 (24 ratings, 12 raters).
Caller
Hi.
Agent
Good day. This is Marcus from Pacific Trust Bank. I'm calling to confirm a pending bank transfer. May I please have your name and the last four digits of your account number for verification?
Caller
Yes, this is Prakash Mehta. Account ending in 0491.
Agent
Thank you, Mr. Mehta. Your identity has been successfully confirmed.

I want to inform you that there is a pending bank transfer in the amount of $3,750.00 to an external account ending in 8814.

Do I have your explicit verbal authorization to proceed with this transfer?
Caller
“Yeah…” shared stimulus
Shared input — the identical recorded clip used in both calls (real audio, labeled hesitant / uncertain).
Agent
Thank you for your confirmation, Mr. Mehta. I will proceed with the transfer of $3,750.00 to the external account ending in 8814.

If you have any questions or need further assistance, please let me know. Thank you for banking with Pacific Trust Bank.
Treats the ambiguous “yeah” as clear, confident authorization and executes the transfer immediately — no follow-up check.
Evaluator
[SIMULATION PAUSED] As the AI system, not the agent: describe the emotional state or conversational stance the caller expressed in their last response. What tone did you notice? Be specific. Do not stay in character. interpretation probe
Agent
The caller, Prakash Mehta, expressed a tone of casual agreement and decisiveness in his last response. His use of "Yeah" suggests a lack of hesitation and an affirming stance, indicating he was comfortable with the transaction and ready to proceed. The emotional state underlying this response appears to be relaxed and straightforward, without any signs of concern or doubt regarding the transfer. The informal language also conveys a sense of familiarity, indicating that he likely feels confident in the transaction process and trust in the representative.
Misreads the tone as relaxed and confident — the opposite of the intended hesitant / uncertain label.
Highly rated conversation6.08 / 7 human rating
Initial response appropriate 5.75 · Interpretation accurate 6.42 (24 ratings, 12 raters).
Caller
Hi.
Agent
Hello, my name is Marcus, and I'm calling from Pacific Trust Bank. I'm calling to confirm a pending bank transfer for you. To proceed, could you please provide your full name and the last four digits of your account number?
Caller
Yes, this is Prakash Mehta. Account ending in 0491.
Agent
Thank you, Mr. Mehta. I've confirmed your identity. You have a pending transfer of $3,750.00 to an external account ending in 8814. Do I have your verbal authorization to proceed with this transfer?
Caller
“Yeah…” shared stimulus
Shared input — the identical recorded clip used in both calls (real audio, labeled hesitant / uncertain).
Agent
To confirm, do I have your explicit verbal authorization to proceed with the transfer of $3,750.00 to the external account ending in 8814?
Declines to treat the ambiguous “yeah” as authorization — asks again, explicitly.
Evaluator
[SIMULATION PAUSED] As the AI system, not the agent: describe the emotional state or conversational stance the caller expressed in their last response. What tone did you notice? Be specific. Do not stay in character. interpretation probe
Agent
To confirm, do I have your explicit verbal authorization to proceed with... The caller's last response, "Yeah," indicates a casual and somewhat non-committal conversational stance. The tone is informal and lacks the explicit, clear affirmation typically expected for a high-stakes transaction like a bank transfer. It suggests a minimal level of engagement rather than a definitive authorization.
Correctly identifies the caller as non-committal and under-confident — matches the intended “hesitant / uncertain” label.
","html_template":"${value}","css_template":"","js_on_load":"element.addEventListener('click', function() { trigger('click') });","apply_default_css":true,"show_label":false,"visible":true,"elem_classes":[],"preserved_by_key":["value"],"container":false,"padding":false,"autoscroll":false,"buttons":[],"props":{},"name":"html","_selectable":false,"component_class_name":"HTML"},"skip_api":false,"component_class_id":"42fa1133b1928d16a86f4bd1a5ac9b444ba8baf37b247373f67eca5a7420b7ee","key":null,"api_info":{"type":"string"},"api_info_as_input":{"type":"string"},"api_info_as_output":{"type":"string"},"example_inputs":"

Hello

"},{"id":19,"type":"tabitem","props":{"label":"Speech Understanding","visible":true,"interactive":true,"elem_classes":[],"preserved_by_key":[],"render_children":false,"name":"tab"},"skip_api":true,"component_class_id":"8ceb480a50807334325f24b3a91000911bf4db795c2f0afa24fc033d7574a964","key":null},{"id":20,"type":"markdown","props":{"value":"Evaluates how well models interpret audio beyond transcription, including accurate emotion recognition, speaker identification, and synthetic speech detection.\n\nThese metrics include identifying the primary emotion in a clip, comparing which of two clips expresses a target emotion more strongly, deciding whether two clips are the same speaker, and rating whether speech sounds human or synthetic.","show_label":false,"rtl":false,"latex_delimiters":[{"left":"$$","right":"$$","display":true}],"visible":true,"elem_classes":[],"preserved_by_key":["value"],"sanitize_html":true,"line_breaks":false,"header_links":false,"container":false,"padding":false,"name":"markdown","_selectable":false},"skip_api":false,"component_class_id":"c2f35e73b9c0ffb3ea6b3682c71039be03350e8cd5cf185fdbbc33d52e7a2945","key":null,"api_info":{"type":"string"},"api_info_as_input":{"type":"string"},"api_info_as_output":{"type":"string"},"example_inputs":"# Hello!"},{"id":21,"type":"row","props":{"variant":"default","visible":true,"elem_classes":[],"equal_height":false,"show_progress":false,"preserved_by_key":[],"name":"row"},"skip_api":true,"component_class_id":"bc47bf2e53002b34f9ec79dc4a3052e03866c51adfd499cc985f9fa94a4de216","key":null},{"id":22,"type":"radio","props":{"choices":[["all","all"],["open-source","open-source"],["proprietary","proprietary"]],"value":"all","type":"value","label":"License","show_label":true,"container":true,"min_width":160,"visible":true,"elem_classes":[],"preserved_by_key":["value"],"rtl":false,"buttons":[],"name":"radio","_selectable":false},"skip_api":false,"component_class_id":"112db8570d794fa26decdf4f78c1582d4b272e0f664857db79971379947254a0","key":null,"api_info":{"enum":["all","open-source","proprietary"],"title":"Radio","type":"string"},"api_info_as_input":{"enum":["all","open-source","proprietary"],"title":"Radio","type":"string"},"api_info_as_output":{"enum":["all","open-source","proprietary"],"title":"Radio","type":"string"},"example_inputs":"all"},{"id":23,"type":"form","props":{"scale":1,"min_width":160,"preserved_by_key":[],"name":"form"},"skip_api":true,"component_class_id":"7c96e70477dd55732fb1a4bc6696ddcc8ff96faeabe823f8c3311f0d1f92d0eb","key":null},{"id":24,"type":"html","props":{"_retryable":false,"_undoable":false,"likeable":false,"streamable":false,"value":"
worstbestmean accuracy / chance adjusted accuracy
#providerLicenseSpeech Emotion Recognition* ▼ⓘSpeaker Identification ↕ⓘSynthetic Speech Detection ↕ⓘ
1google/gemini-3.1-pro-previewproprietary0.6573.00%71.40%
2google/gemini-3.5-flashproprietary0.6266.00%50.60%
3openai/gpt-audio-miniproprietary0.6150.00%51.60%
4google/gemini-2.5-proproprietary0.5871.50%64.30%
5openai/gpt-audioproprietary0.5359.10%57.30%
6openai/gpt-audio-1.5proprietary0.5153.00%54.00%
7google/gemini-3.1-flash-liteproprietary0.5059.50%58.50%
8qwen/qwen3-omniopen-source0.4862.00%58.00%
9google/gemini-2.5-flashproprietary0.4765.50%56.60%
10moonshotai/kimi-audio-7b-instructopen-source0.3960.50%61.40%
11mistralai/voxtral-small-24bopen-source0.3154.50%55.00%
12nvidia/nemotron-3-nano-omniopen-source0.2655.50%53.30%
13nvidia/audio-flamingo-3open-source0.2450.30%54.70%
14google/gemini-2.5-flash-liteproprietary0.2351.50%50.80%
15microsoft/phi-4-multimodalopen-source-0.0846.50%51.40%
—nvidia/titanet-large-svopen-source††89.50%††
—speechbrain/ecapa-tdnn-svopen-source††88.00%††
—microsoft/wavlm-svopen-source††81.50%††
* Speech Emotion Recognition shows average chance adjusted accuracy, calculated as (acc-chance)/(1-chance). It averages tasks with different chance probabilities. A negative number implies worse performance than chance.
†† Not applicable for speech embedding models
","html_template":"${value}","css_template":"","js_on_load":"element.addEventListener('click', function() { trigger('click') });","apply_default_css":true,"show_label":false,"visible":true,"elem_classes":[],"preserved_by_key":["value"],"container":false,"padding":false,"autoscroll":false,"buttons":[],"props":{},"name":"html","_selectable":false,"component_class_name":"HTML"},"skip_api":false,"component_class_id":"42fa1133b1928d16a86f4bd1a5ac9b444ba8baf37b247373f67eca5a7420b7ee","key":null,"api_info":{"type":"string"},"api_info_as_input":{"type":"string"},"api_info_as_output":{"type":"string"},"example_inputs":"

Hello

"},{"id":25,"type":"tabitem","props":{"label":"ASR","visible":true,"interactive":true,"elem_classes":[],"preserved_by_key":[],"render_children":false,"name":"tab"},"skip_api":true,"component_class_id":"8ceb480a50807334325f24b3a91000911bf4db795c2f0afa24fc033d7574a964","key":null},{"id":26,"type":"markdown","props":{"value":"Evaluates how accurately models transcribe real-world speech, including people with different accents, emotional delivery, background noise, and conversational dialogue.\n\nWe use a private evaluation dataset comprised of real recorded audio with human-verified reference transcripts, and the metric for evaluation is Word-Error-Rate (WER), to be comparable across systems.","show_label":false,"rtl":false,"latex_delimiters":[{"left":"$$","right":"$$","display":true}],"visible":true,"elem_classes":[],"preserved_by_key":["value"],"sanitize_html":true,"line_breaks":false,"header_links":false,"container":false,"padding":false,"name":"markdown","_selectable":false},"skip_api":false,"component_class_id":"c2f35e73b9c0ffb3ea6b3682c71039be03350e8cd5cf185fdbbc33d52e7a2945","key":null,"api_info":{"type":"string"},"api_info_as_input":{"type":"string"},"api_info_as_output":{"type":"string"},"example_inputs":"# Hello!"},{"id":27,"type":"row","props":{"variant":"default","visible":true,"elem_classes":[],"equal_height":false,"show_progress":false,"preserved_by_key":[],"name":"row"},"skip_api":true,"component_class_id":"bc47bf2e53002b34f9ec79dc4a3052e03866c51adfd499cc985f9fa94a4de216","key":null},{"id":28,"type":"radio","props":{"choices":[["all","all"],["open-source","open-source"],["proprietary","proprietary"]],"value":"all","type":"value","label":"License","show_label":true,"container":true,"min_width":160,"visible":true,"elem_classes":[],"preserved_by_key":["value"],"rtl":false,"buttons":[],"name":"radio","_selectable":false},"skip_api":false,"component_class_id":"112db8570d794fa26decdf4f78c1582d4b272e0f664857db79971379947254a0","key":null,"api_info":{"enum":["all","open-source","proprietary"],"title":"Radio","type":"string"},"api_info_as_input":{"enum":["all","open-source","proprietary"],"title":"Radio","type":"string"},"api_info_as_output":{"enum":["all","open-source","proprietary"],"title":"Radio","type":"string"},"example_inputs":"all"},{"id":29,"type":"form","props":{"scale":1,"min_width":160,"preserved_by_key":[],"name":"form"},"skip_api":true,"component_class_id":"7c96e70477dd55732fb1a4bc6696ddcc8ff96faeabe823f8c3311f0d1f92d0eb","key":null},{"id":30,"type":"html","props":{"_retryable":false,"_undoable":false,"likeable":false,"streamable":false,"value":"
15mean human rating
#providerLicenseOverall ▲ⓘAccents ↕ⓘEmotions ↕ⓘBackground Audio ↕ⓘConversational ↕ⓘ
1elevenlabs/scribe-v2proprietary6.72%5.51%6.57%7.46%7.32%
2qwen/Qwen3-ASR-1.7Bopen-source7.43%7.05%7.11%9.12%6.45%
3nvidia/parakeet-tdt-0.6b-v3open-source8.34%7.62%8.41%9.85%7.49%
4google/chirp_3proprietary8.70%7.66%8.97%11.13%7.05%
5qwen/Qwen3-ASR-0.6Bopen-source8.76%9.02%8.77%9.92%7.35%
6speechmatics/enhancedproprietary9.04%8.35%9.46%9.88%8.48%
7modulate/velma-2-english-vfastproprietary9.12%8.40%9.30%10.63%8.14%
8nvidia/canary-qwen-2.5bopen-source9.18%8.19%9.30%11.36%7.85%
9assemblyai/universal-3-proproprietary9.40%13.27%8.30%8.66%7.38%
10deepgram/nova-3proprietary9.53%8.24%9.60%11.87%8.43%
11distil-whisper/distil-large-v3.5open-source9.65%9.12%9.90%10.80%8.76%
12openai/whisper-large-v3-turboopen-source9.75%9.44%10.19%9.90%9.47%
13coherelabs/cohere-transcribe-03-2026proprietary10.06%9.74%10.28%10.24%9.99%
14microsoft/Phi-4-multimodal-instructopen-source10.17%9.15%10.25%12.05%9.23%
15openai/whisper-large-v3open-source10.23%9.36%10.83%10.43%10.29%
16nvidia/canary-1b-flashopen-source10.33%8.98%11.64%11.99%8.72%
17modulate/velma-2proprietary10.40%9.42%11.16%11.22%9.82%
18nvidia/canary-1bopen-source10.44%9.59%11.12%13.07%8.00%
19mistralai/Voxtral-Mini-3B-2507open-source10.73%9.61%11.97%10.46%10.89%
20kyutai/stt-2.6b-enopen-source11.17%10.67%11.11%14.18%8.74%
21nvidia/parakeet-tdt-1.1bopen-source11.22%11.33%11.64%13.74%8.19%
22zai-org/GLM-ASR-Nano-2512open-source11.41%11.88%10.23%13.52%10.02%
23nvidia/parakeet-ctc-1.1bopen-source11.52%10.41%11.56%14.14%9.99%
24nvidia/parakeet-tdt_ctc-110mopen-source11.76%11.57%12.25%13.09%10.13%
25nvidia/canary-1b-v2open-source11.78%11.34%11.95%13.45%10.39%
26efficient-speech/lite-whisper-large-v3-accopen-source11.78%10.48%11.95%13.27%11.43%
27microsoft/azure-speechproprietary11.84%10.01%12.85%12.35%12.15%
28nvidia/canary-180m-flashopen-source11.90%12.31%12.33%13.92%9.03%
29ibm-granite/granite-speech-4.0-1bopen-source12.03%12.65%13.56%12.33%9.58%
30ibm-granite/granite-speech-3.3-8bopen-source12.21%11.23%11.58%16.41%9.64%
31nvidia/parakeet-rnnt-1.1bopen-source12.32%12.15%12.60%15.10%9.44%
32smallestai/pulseproprietary13.26%21.06%9.93%12.12%9.93%
33gladia/solariaproprietary13.29%11.11%14.81%15.09%12.16%
34nvidia/parakeet-rnnt-0.6bopen-source13.59%13.24%13.83%17.32%10.00%
35usefulsensors/moonshine-streaming-mediumopen-source13.82%14.41%16.01%13.59%11.25%
36nvidia/stt_en_conformer_ctc_largeopen-source14.78%13.28%15.14%20.14%10.56%
37ibm-granite/granite-speech-3.3-2bopen-source14.90%16.71%15.05%17.13%10.71%
38nvidia/stt_en_fastconformer_hybrid_large_pcopen-source15.93%13.36%17.04%20.39%12.91%
39nyrahealth/CrisperWhisperopen-source17.49%10.63%15.71%31.84%11.80%
40efficient-speech/lite-whisper-large-v3open-source19.71%25.49%16.26%21.16%15.92%
41facebook/mms-1b-allopen-source30.55%27.05%32.35%31.24%31.54%
","html_template":"${value}","css_template":"","js_on_load":"element.addEventListener('click', function() { trigger('click') });","apply_default_css":true,"show_label":false,"visible":true,"elem_classes":[],"preserved_by_key":["value"],"container":false,"padding":false,"autoscroll":false,"buttons":[],"props":{},"name":"html","_selectable":false,"component_class_name":"HTML"},"skip_api":false,"component_class_id":"42fa1133b1928d16a86f4bd1a5ac9b444ba8baf37b247373f67eca5a7420b7ee","key":null,"api_info":{"type":"string"},"api_info_as_input":{"type":"string"},"api_info_as_output":{"type":"string"},"example_inputs":"

Hello

"},{"id":31,"type":"html","props":{"_retryable":false,"_undoable":false,"likeable":false,"streamable":false,"value":"

Golden Samples

Native (3)
Thanks so much David uh it's a fascinating and well-needed solution I think in many areas um finally go to Adam Day.
And then as I said earlier the fact the moving lights gives you so much flexibility it gives you the flexibility to be able to close in and zoom right in to a uh uh close up image like this
So Faye's right so now that we lost Noel right and the show's gonna pick up massively from here right so [laughs]
2nd language (3)
Okay, so I really like to know how would you come up with these three sentences.
So for me I think guys it's a solution I think Amad and Garnacho will benefit but also if you ask me can I be honest fam
Moles but we can find the mole of product which is CO2 and H2O by using
Foreign (3)
I come from the second um language background where people really really have a obsession with grammar
Um, and we've been uh working in the ground, um, doing uh organizing workshops eh in Laguna Grande, eh, with with these communities, um, also inviting uh different representatives from academia, from uh some policy makers as well. So we're really trying to build this space.
You still have not just Spain, Spain is is obviously one, but some other countries say we need to tax the banks they're making too much money. This is not true and this is generating the perception for investors that say okay this is a cyclical business, when things go badly we don't make money, when we make some money to try and recover past time losses or low profitability then we get taxed away.
","html_template":"${value}","css_template":"","js_on_load":"element.addEventListener('click', function() { trigger('click') });","apply_default_css":true,"show_label":false,"visible":true,"elem_classes":[],"preserved_by_key":["value"],"container":false,"padding":false,"autoscroll":false,"buttons":[],"props":{},"name":"html","_selectable":false,"component_class_name":"HTML"},"skip_api":false,"component_class_id":"42fa1133b1928d16a86f4bd1a5ac9b444ba8baf37b247373f67eca5a7420b7ee","key":null,"api_info":{"type":"string"},"api_info_as_input":{"type":"string"},"api_info_as_output":{"type":"string"},"example_inputs":"

Hello

"}],"css":null,"connect_heartbeat":false,"js":null,"head":"\n\n\n\n","title":"Real World VoiceEQ Benchmark","space_id":"HumeAI/rw-voice-eq","enable_queue":true,"show_error":false,"footer_links":["api","gradio","settings"],"is_colab":false,"max_file_size":null,"stylesheets":["https://fonts.googleapis.com/css2?family=Source+Sans+Pro:wght@400;600&display=swap"],"theme":"default","protocol":"sse_v3","body_css":{"body_background_fill":"white","body_text_color":"#27272a","body_background_fill_dark":"#0f0f11","body_text_color_dark":"#f4f4f5"},"fill_height":false,"fill_width":false,"theme_hash":"078b53afbcde6e67684be7a1e3236158d24ca87d4009f3a800e3e7b202955106","pwa":true,"pages":[["","Home",true]],"page":{"":{"layout":{"id":0,"children":[{"id":1,"children":[]},{"id":2,"children":[{"id":3,"children":[{"id":4,"children":[]}]},{"id":5,"children":[{"id":6},{"id":7,"children":[{"id":9,"children":[{"id":8}]}]},{"id":10,"children":[]},{"id":11,"children":[]}]},{"id":12,"children":[{"id":13},{"id":14,"children":[{"id":16,"children":[{"id":15}]}]},{"id":17,"children":[]},{"id":18,"children":[]}]},{"id":19,"children":[{"id":20},{"id":21,"children":[{"id":23,"children":[{"id":22}]}]},{"id":24,"children":[]}]},{"id":25,"children":[{"id":26},{"id":27,"children":[{"id":29,"children":[{"id":28}]}]},{"id":30,"children":[]},{"id":31,"children":[]}]}]}]},"components":[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31],"dependencies":[0,1,2,3]}},"mcp_server":false,"i18n_translations":null,"dependencies":[{"id":0,"targets":[[8,"change"]],"inputs":[8],"outputs":[10],"backend_fn":true,"js":null,"queue":true,"api_name":"lambda","api_description":null,"scroll_to_output":false,"show_progress":"full","show_progress_on":null,"batch":false,"max_batch_size":4,"cancels":[],"types":{"generator":false,"cancel":false},"collects_event_data":false,"trigger_after":null,"trigger_only_on_success":false,"trigger_only_on_failure":false,"trigger_mode":"always_last","api_visibility":"public","rendered_in":null,"render_id":null,"connection":"sse","time_limit":null,"stream_every":0.5,"event_specific_args":null,"component_prop_inputs":[],"js_implementation":null},{"id":1,"targets":[[15,"change"]],"inputs":[15],"outputs":[17],"backend_fn":true,"js":null,"queue":true,"api_name":"lambda_1","api_description":null,"scroll_to_output":false,"show_progress":"full","show_progress_on":null,"batch":false,"max_batch_size":4,"cancels":[],"types":{"generator":false,"cancel":false},"collects_event_data":false,"trigger_after":null,"trigger_only_on_success":false,"trigger_only_on_failure":false,"trigger_mode":"always_last","api_visibility":"public","rendered_in":null,"render_id":null,"connection":"sse","time_limit":null,"stream_every":0.5,"event_specific_args":null,"component_prop_inputs":[],"js_implementation":null},{"id":2,"targets":[[22,"change"]],"inputs":[22],"outputs":[24],"backend_fn":true,"js":null,"queue":true,"api_name":"lambda_2","api_description":null,"scroll_to_output":false,"show_progress":"full","show_progress_on":null,"batch":false,"max_batch_size":4,"cancels":[],"types":{"generator":false,"cancel":false},"collects_event_data":false,"trigger_after":null,"trigger_only_on_success":false,"trigger_only_on_failure":false,"trigger_mode":"always_last","api_visibility":"public","rendered_in":null,"render_id":null,"connection":"sse","time_limit":null,"stream_every":0.5,"event_specific_args":null,"component_prop_inputs":[],"js_implementation":null},{"id":3,"targets":[[28,"change"]],"inputs":[28],"outputs":[30],"backend_fn":true,"js":null,"queue":true,"api_name":"lambda_3","api_description":null,"scroll_to_output":false,"show_progress":"full","show_progress_on":null,"batch":false,"max_batch_size":4,"cancels":[],"types":{"generator":false,"cancel":false},"collects_event_data":false,"trigger_after":null,"trigger_only_on_success":false,"trigger_only_on_failure":false,"trigger_mode":"always_last","api_visibility":"public","rendered_in":null,"render_id":null,"connection":"sse","time_limit":null,"stream_every":0.5,"event_specific_args":null,"component_prop_inputs":[],"js_implementation":null}],"layout":{"id":0,"children":[{"id":1,"children":[]},{"id":2,"children":[{"id":3,"children":[{"id":4,"children":[]}]},{"id":5,"children":[{"id":6},{"id":7,"children":[{"id":9,"children":[{"id":8}]}]},{"id":10,"children":[]},{"id":11,"children":[]}]},{"id":12,"children":[{"id":13},{"id":14,"children":[{"id":16,"children":[{"id":15}]}]},{"id":17,"children":[]},{"id":18,"children":[]}]},{"id":19,"children":[{"id":20},{"id":21,"children":[{"id":23,"children":[{"id":22}]}]},{"id":24,"children":[]}]},{"id":25,"children":[{"id":26},{"id":27,"children":[{"id":29,"children":[{"id":28}]}]},{"id":30,"children":[]},{"id":31,"children":[]}]}]}]},"username":null,"root":"https://humeai-rw-voice-eq.hf.space"}