Fish AudioAudio modelActive
Fish Audio S2.1 Pro
PopularityRank 670 of 875
0out of 100same as last month
Summary
Fish Audio S2.1 Pro is the API endpoint for the newer S2.1 generation. It produces multilingual speech, supports voice cloning, and responds in about 90 milliseconds to first audio.
The free identifier s2.1-pro-free is announced through November 30, 2026 under fair-use rules and without an SLA. The offer’s terms govern data retention and commercial use. No separate S2.1 weights checkpoint is published.
All benchmark results
Every published result with its source, test setup and comparison models.
Show all 183 published benchmark results
Page 1 of 16
Measurements without matching peer values
These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.
win rate versus S2 Pro (Show measurement, test conditions, and source)
- Source value
- 61
- Score
- 61
- Metric
- win rate
- Unit
- %
- Benchmark version
- source version not stated
- Category
- audio
- Direction
- Higher is better
- Source type
- vendor-reported
- Evaluator
- Fish Audio S2.1 announcement
- Status
- active
- Retrieved at
- 2026-10-04
- Methodology
- Vendor-reported result. Provider blind listening; exact sample count not stated in announcement
- Context
- Source-specific observation; it does not establish a compatible cross-model cohort.
Coval rolling-30-day pooled output word error rate (s2.1-pro-free endpoint) (Show measurement, test conditions, and source)
- Source value
- 1.5362004150695352
- Score
- 1.5362004150695352
- Metric
- Pooled word error rate of speech synthesized by the exact free endpoint
- Unit
- %
- Benchmark version
- Coval rolling 30-day snapshot, last measured 2026-10-16, source response fetched 2026-10-16T15:36:04.055765Z
- Category
- audio
- Direction
- Lower is better
- Sample
- 2097
- Source type
- independent-evaluator
- Evaluator
- Coval
- Status
- superseded
- Retrieved at
- 2026-10-16
- Methodology
- Independent evaluation. Exact Fish Audio API model ID s2.1-pro-free, corresponding to the public free service route. Coval measures that route separately from production S2.1 Pro, using the same prompts, and reports a rolling 30-day result last measured 2026-10-16. Stored score is pooled_value=1.5362004150695352 from the WER row, not mean_value; Coval's visible 1.5% is rounded to one decimal. The reported WER sample count comes from the same row.
- Context
- This score is for the free s2.1-pro-free endpoint, not the separately measured paid production S2.1 Pro endpoint. The reported WER sample count covers scored items only, not all attempted requests or failures. No confidence interval is reported; free-tier availability or policy can change. TTS WER checks intelligibility, not naturalness or human preference.
Coval rolling-30-day TTS word error rate, 2026-10-17 snapshot (Show measurement, test conditions, and source)
- Source value
- 1.5
- Score
- 1.5
- Metric
- Coval-reported TTS word error rate (WER)
- Unit
- %
- Benchmark version
- Coval daily fixed-prompt endpoint benchmark, rolling 30-day window; route last measured 2026-10-17
- Category
- audio
- Direction
- Lower is better
- Sample
- 2160
- Source type
- independent-evaluator
- Evaluator
- Coval Voice AI Benchmarks
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Coval exact hosted model route S2.1 Pro by Fish Audio, not the separate S2.1 Pro Free route. Rolling 30-day fixed-prompt TTS evaluation. The page reports 1.5% WER on 2160 samples for the route last measured 2026-10-17.
- Context
- This is the hosted production route and must not be merged with the separately named free route. Coval's visible WER is rounded to one decimal place; no confidence interval is shown.
Hume Voice Controllability role-fit win rate (Show measurement, test conditions, and source)
- Source value
- 33
- Score
- 33
- Metric
- Blind pairwise listener win rate over decided voice-role-fit votes
- Unit
- %
- Benchmark version
- Hume Voice Controllability leaderboard, overall role-fit table, captured 2026-10-17
- Category
- audio
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Hume AI Real World VoiceEQ
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Hume exact table label fish-speech/s2.1-pro. Paid blind raters compare two voices made for the same production brief. Score is the share of decided votes for role fit; 50% is even.
- Context
- This is a pairwise role-fit win rate, not a direct naturalness rating. Hume does not publish a per-model vote count or confidence interval in the table.
Hume Voice Controllability naturalness win rate (Show measurement, test conditions, and source)
- Source value
- 34.6
- Score
- 34.6
- Metric
- Blind pairwise listener win rate over decided naturalness votes
- Unit
- %
- Benchmark version
- Hume Voice Controllability leaderboard, overall role-fit table, captured 2026-10-17
- Category
- audio
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Hume AI Real World VoiceEQ
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Hume exact table label fish-speech/s2.1-pro. Paid blind raters separately judge which voice sounds more natural for the same production brief. Score is the share of decided votes; 50% is even.
- Context
- This is a pairwise naturalness win rate and is distinct from role fit. Hume does not publish a per-model vote count or confidence interval in the table.
Hume Voice Design age-band accuracy (Show measurement, test conditions, and source)
- Source value
- 56
- Score
- 56
- Metric
- Accuracy identifying the requested generated-voice age band
- Unit
- %
- Benchmark version
- Hume Voice Controllability leaderboard, age table, captured 2026-10-17
- Category
- audio
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Hume AI Real World VoiceEQ
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Hume exact table label fish-speech/s2.1-pro. Listeners choose among age bands 20s, 40s and 60s; the source gives 33% as chance accuracy.
- Context
- This evaluates recognition of the requested age band, not age estimation of a real speaker. Per-model N and confidence intervals are not published.
Hume Voice Design gender consistency accuracy (Show measurement, test conditions, and source)
- Source value
- 99.1
- Score
- 99.1
- Metric
- Accuracy identifying the intended generated-voice gender
- Unit
- %
- Benchmark version
- Hume Voice Controllability leaderboard, gender table, captured 2026-10-17
- Category
- audio
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Hume AI Real World VoiceEQ
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Hume exact table label fish-speech/s2.1-pro. Listeners identify the gender the voice was built as; the source gives 50% as chance accuracy.
- Context
- This is one generated-voice control dimension, not an overall TTS quality score. Per-model N and confidence intervals are not published.
Hume Voice Design requested-texture match accuracy (Show measurement, test conditions, and source)
- Source value
- 60.8
- Score
- 60.8
- Metric
- Accuracy selecting the requested vocal texture in a two-voice comparison
- Unit
- %
- Benchmark version
- Hume Voice Controllability leaderboard, texture table, captured 2026-10-17
- Category
- audio
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Hume AI Real World VoiceEQ
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Hume exact table label fish-speech/s2.1-pro. Raters choose which of two voices better matches a requested texture word; the source gives 50% as chance accuracy.
- Context
- The aggregate covers texture words such as nasal, breathy, raspy, warm, resonant and bright. Per-model N and confidence intervals are not published.
Hume Voice Design accent identification accuracy (Show measurement, test conditions, and source)
- Source value
- 25.7
- Score
- 25.7
- Metric
- Accuracy pooled over six voice-design accent tasks
- Unit
- %
- Benchmark version
- Hume Voice Controllability leaderboard, accent table, captured 2026-10-17
- Category
- audio
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Hume AI Real World VoiceEQ
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Hume exact table label fish-speech/s2.1-pro. The table pools six accent tasks; the source states that chance varies between 17% and 33% by task.
- Context
- This pooled value has task-specific chance levels from 17% to 33%, so it should not be read as a single-task accuracy or compared directly with the other Hume tabs. Per-model N and confidence intervals are not published.
Hume voice-dial pass rate across four controls (Show measurement, test conditions, and source)
- Source value
- 74.3
- Score
- 74.3
- Metric
- Share of audio pairs audibly moving in the intended direction, averaged across four dials
- Unit
- %
- Benchmark version
- Hume Voice Controllability leaderboard, dial table, captured 2026-10-17
- Category
- audio
- Direction
- Higher is better
- Source type
- independent-evaluator
- Evaluator
- Hume AI Real World VoiceEQ
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Hume exact table label fish-speech/s2.1-pro. The four controls are pitch, tempo, volume and pitch range. This result is measured directly from paired audio and does not use human raters.
- Context
- This measures controllability of the requested dial direction, not naturalness or speech quality. The source does not disclose per-model N or a confidence interval.
Coval Fish Audio S2.1 Pro TTS WER, run 19:00 UTC (Show measurement, test conditions, and source)
- Source value
- 1.6
- Score
- 1.6
- Metric
- Coval fixed-prompt speech-intelligibility word error rate, displayed average
- Unit
- %
- Benchmark version
- Coval daily public evaluator run; rolling 30-day window; last measured 2026-10-17 19:00 UTC
- Category
- audio
- Direction
- Lower is better
- Sample
- 2216
- Source type
- independent-evaluator
- Evaluator
- Coval Voice AI Benchmarks
- Status
- active
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Coval public fixed-prompt evaluation, rolling 30-day window and daily run. Exact page route/model label: Fish Audio S2.1 Pro production route; created by Fish Audio, hosted by Fish Audio. Reported values and metric-specific denominators are retained; milliseconds are converted exactly to seconds. Last run stated on page: 2026-10-17 19:00 UTC.
- Context
- Coval publishes rolling 30-day results from daily repeated runs on fixed public audio or prompt sets. These rows are the later timestamped run visible in the captured page, not a copy of the earlier run in the input. Exact API/checkpoint revision, request parameters, and confidence intervals are not published; displayed WER is rounded to one decimal and is not a measure of naturalness.
Coval Fish Audio S2.1 Pro TTS WER p95, run 19:00 UTC (Show measurement, test conditions, and source)
- Source value
- 12.2
- Score
- 12.2
- Metric
- 95th-percentile per-prompt TTS word error rate
- Unit
- %
- Benchmark version
- Coval daily public evaluator run; rolling 30-day window; last measured 2026-10-17 19:00 UTC
- Category
- audio
- Direction
- Lower is better
- Sample
- 2216
- Source type
- independent-evaluator
- Evaluator
- Coval Voice AI Benchmarks
- Status
- superseded
- Retrieved at
- 2026-10-17
- Methodology
- Independent evaluation. Coval public fixed-prompt evaluation, rolling 30-day window and daily run. Exact page route/model label: Fish Audio S2.1 Pro production route; created by Fish Audio, hosted by Fish Audio. Reported values and metric-specific denominators are retained; milliseconds are converted exactly to seconds. Last run stated on page: 2026-10-17 19:00 UTC.
- Context
- Coval publishes rolling 30-day results from daily repeated runs on fixed public audio or prompt sets. These rows are the later timestamped run visible in the captured page, not a copy of the earlier run in the input. Exact API/checkpoint revision, request parameters, and confidence intervals are not published; displayed WER is rounded to one decimal and is not a measure of naturalness.
Page 1 of 16
Popularity compared
How often the model is searched for, used and mentioned, compared with every model of its category.
Popularity index
0 to 100, higher is more popular
20 of 83 models
pyannote Community-1, pyannote: 73
Nemotron-3-Diarization, NVIDIA: 57
Gemini 3.5 Transcribe, Google: 52
Confucius4-R2T2, NetEase Youdao: 52
Granite Speech 5.0 470M TurboCTC, IBM: 52
Kokoro-82M, hexgrad: 51
Nova-3, Deepgram: 51
Fish Audio S2 Pro, Fish Audio: 51
NVIDIA Nemotron 3.5 Transcribe, NVIDIA: 50
Granite Speech 5.0 470M TurboCTC NC, IBM: 50
Orukeet, Oruk: 49
Lyria 3.5, Google: 47
Gemini 3.5 Live Translate, Google: 46
Phonon-2, Fermion Research: 44
Aura-2, Deepgram: 44
Gemini 3.1 Flash TTS, Google: 44
MiniMax-Music3, MiniMax: 41
Audio8-ASR-Infinite, Edge0 AI: 41
GPT-Realtime 2.1, OpenAI: 41
Fish Audio S2.1 Pro, Fish Audio: 0
Fish Audio S2.1 Pro
As of November 2026 · Source: gradually.ai popularity index
Over time
Weekly values from October 18, 2026 to November 5, 2026
Fish Audio S2.1 ProAmazon Nova 2 SonicEleven Multilingual STS v2Eleven Music v2.5Eleven Text to Sound v2
All specifications
Architecture
Input and output
Watermarking and provenance
AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.
- AI watermarking
- Unknown
- no source
- Watermarking technique
- Not known
- no source
Licence and access
- Usage conditions
- No SLA or guaranteed latency; possible retention and commercial-use restrictions apply
- Source: Usage conditions
API
Prices
- Free API access
- Free through 2026-11-30 under fair-use rules; changes with advance noticeSource
Sources and data date
Show all 45 sources
Benchmark sources at a glance
Which independent evaluators tested this model. This is coverage, not a quality ranking.
No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.
Without vendor claims and historical or disputed results; those stay labelled in the single values.