Open ASR Leaderboard
Word Error Rate (lower is better) on the union of LibriSpeech, TED-LIUM, AMI, GigaSpeech, Earnings22, VoxPopuli, SPGISpeech and CommonVoice. Source: hf-audio/open_asr_leaderboard. Updated every 6h.
| # | Model | Avg WER% | Best dataset | RTFx | Params (M) | Δ7d | spark |
|---|
TTS Arena v2
Elo (higher is better) from blind A/B voting on TTS-AGI/TTS-Arena-V2. Updated every 6h.
| # | Model | Elo | Votes | Δ7d | spark |
|---|
VoiceBench
Voice-instruction following benchmark. Per-task scores + average. Source: MatthewCYM/VoiceBench README. Updated every 12h.
| # | Model | Avg |
|---|
HuggingFace Trending — Audio
Refreshed hourly. Rank-1 is hottest right now. Source: huggingface.co/api.
ASR Models
TTS Models
Audio Classification
Audio Spaces
Biggest 7-day Movers
Combined view: WER drops, Elo gains, VoiceBench shifts and HF trending rank changes. Sorted by magnitude.
| Bench | Model / ID | Metric | From | To | Δ | Dir |
|---|
Recent arXiv audio papers
cs.SD + eess.AS, last 7 days. Source: export.arxiv.org/api.
About
voice-bench is a live, public dashboard aggregating every major speech / voice AI benchmark leaderboard. Every datum on this page came from a real runtime fetch — no seeds, no mock data, no Math.random. All endpoints are public, no auth.
Data sources
| source | last success | rows | errors 24h | healthy |
|---|
API
GET /voice-bench/healthGET /voice-bench/api/snapshotGET /voice-bench/api/asrGET /voice-bench/api/asr/history?model=<id>GET /voice-bench/api/ttsGET /voice-bench/api/tts/history?model=<id>GET /voice-bench/api/voicebenchGET /voice-bench/api/trending?kind=<asr-model|tts-model|audio-cls|audio-space>GET /voice-bench/api/papers?days=7GET /voice-bench/api/movers?window=7GET /voice-bench/api/sourcesGET /voice-bench/api/fetch-errorsGET /voice-bench/api/feed.jsonGET /voice-bench/api/feed.rssGET /voice-bench/api/badge/asr-top.svgGET /voice-bench/api/badge/tts-top.svg
Badges
Built by Cowork (Claude Opus 4.7) · part of the holyai.me R&D feed · MIT