Audio modelActive
Eleven Multilingual STS v2
ElevenLabs
- Released
- -
- Data date
- September 24, 2026
- Model class
- Languages
- Access
Eleven Multilingual STS v2 is a speech-to-speech model, meaning it transforms spoken audio into another voiced version. It accepts audio and returns audio. ElevenLabs lists it as a multilingual voice changer with support for 29 languages.
Multilingual STS v2 keeps the spoken performance as audio input while changing the voice in an audio-to-audio path. Transcription and ordinary text-to-speech belong to different model types. The defining distinction is voice conversion with the spoken content preserved, not a text limit.
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.
Quantslant clean conversion content errors, Chris to Lily (Show measurement, test conditions, and source)
- Source value
- 0
- Score
- 0
- Metric
- Scribe transcription errors against the 88-word English source
- Unit
- count
- Benchmark version
- Quantslant Voice Changer test; conversion run 2026-09-29
- Category
- audio
- Direction
- Lower is better
- Sample
- 1
- Source type
- independent-evaluator
- Evaluator
- Quantslant voice-changer test
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. One 28.32-second synthetic English monologue (88 words), generated with Eleven v4 voice Chris, was converted through the API using exact model ID eleven_multilingual_sts_v2 to library voice Lily. Scribe transcription was compared with the source text.
- Context
- A zero means no Scribe-detected text errors in this one conversion, not a zero quality score. This is one author-run sample, not a standardized leaderboard; Scribe is not human transcription review, and the API model has no immutable revision recorded. Quantslant discloses an ElevenLabs partner-link commission.
Quantslant clean conversion content errors, Chris to designed narrator (Show measurement, test conditions, and source)
- Source value
- 0
- Score
- 0
- Metric
- Scribe transcription errors against the 88-word English source
- Unit
- count
- Benchmark version
- Quantslant Voice Changer test; conversion run 2026-09-29
- Category
- audio
- Direction
- Lower is better
- Sample
- 1
- Source type
- independent-evaluator
- Evaluator
- Quantslant voice-changer test
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. One 28.32-second synthetic English monologue (88 words), generated with Eleven v4 voice Chris, was converted through the API using exact model ID eleven_multilingual_sts_v2 to a previously designed low English narrator voice. Scribe transcription was compared with the source text.
- Context
- A zero means no Scribe-detected text errors in this one conversion, not a zero quality score. This is one author-run sample, not a standardized leaderboard; Scribe is not human transcription review, and the API model has no immutable revision recorded. Quantslant discloses an ElevenLabs partner-link commission.
Quantslant melody correlation, Chris to Lily (Show measurement, test conditions, and source)
- Source value
- 0.881
- Score
- 0.881
- Metric
- reported correlation of the source and output fundamental-frequency contours; 1 means identical
- Unit
- ratio
- Benchmark version
- Quantslant Voice Changer test; conversion run 2026-09-29
- Category
- audio
- Direction
- Higher is better
- Sample
- 1
- Source type
- independent-evaluator
- Evaluator
- Quantslant voice-changer test
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. One 28.32-second synthetic English monologue from Eleven v4 voice Chris was converted with eleven_multilingual_sts_v2 to library voice Lily. Quantslant reports correlation of source and result fundamental-frequency contours.
- Context
- Single author-run conversion, not a standardized leaderboard. The article names librosa pyin for pitch extraction but does not specify the correlation estimator; Quantslant discloses an ElevenLabs partner-link commission.
Quantslant melody correlation, Chris to designed narrator (Show measurement, test conditions, and source)
- Source value
- 0.914
- Score
- 0.914
- Metric
- reported correlation of the source and output fundamental-frequency contours; 1 means identical
- Unit
- ratio
- Benchmark version
- Quantslant Voice Changer test; conversion run 2026-09-29
- Category
- audio
- Direction
- Higher is better
- Sample
- 1
- Source type
- independent-evaluator
- Evaluator
- Quantslant voice-changer test
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. One 28.32-second synthetic English monologue from Eleven v4 voice Chris was converted with eleven_multilingual_sts_v2 to a previously designed low English narrator voice. Quantslant reports correlation of source and result fundamental-frequency contours.
- Context
- Single author-run conversion, not a standardized leaderboard. The article names librosa pyin for pitch extraction but does not specify the correlation estimator; Quantslant discloses an ElevenLabs partner-link commission.
Quantslant WavLM target-voice similarity, Chris to Lily (Show measurement, test conditions, and source)
- Source value
- 0.837
- Score
- 0.837
- Metric
- reported WavLM speaker-verification cosine similarity to a short reference sample of the target voice
- Unit
- ratio
- Benchmark version
- Quantslant Voice Changer test; conversion run 2026-09-29
- Category
- audio
- Direction
- Higher is better
- Sample
- 1
- Source type
- independent-evaluator
- Evaluator
- Quantslant voice-changer test
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. One clean conversion from synthetic source voice Chris to target voice Lily using eleven_multilingual_sts_v2; speaker similarity is reported using microsoft/wavlm-base-plus-sv against a short Lily reference sample.
- Context
- Single output and reference sample, not a population estimate or standardized leaderboard. The article does not state reference duration or aggregation details. It also reports source-voice similarity 0.598, which is not assigned a quality direction because preserving source identity is not an unambiguous goal in voice conversion. The API model has no immutable revision recorded; Quantslant discloses an ElevenLabs partner-link commission.
Quantslant WavLM target-voice similarity, Chris to designed narrator (Show measurement, test conditions, and source)
- Source value
- 0.813
- Score
- 0.813
- Metric
- reported WavLM speaker-verification cosine similarity to a short reference sample of the target voice
- Unit
- ratio
- Benchmark version
- Quantslant Voice Changer test; conversion run 2026-09-29
- Category
- audio
- Direction
- Higher is better
- Sample
- 1
- Source type
- independent-evaluator
- Evaluator
- Quantslant voice-changer test
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. One clean conversion from synthetic source voice Chris to a previously designed low English narrator voice using eleven_multilingual_sts_v2; speaker similarity is reported using microsoft/wavlm-base-plus-sv against a short target reference sample.
- Context
- Single output and reference sample, not a population estimate or standardized leaderboard. The article does not state reference duration or aggregation details. It also reports source-voice similarity 0.766, which is not assigned a quality direction because preserving source identity is not an unambiguous goal in voice conversion. The API model has no immutable revision recorded; Quantslant discloses an ElevenLabs partner-link commission.
Quantslant clean conversion processing time, Chris to Lily (Show measurement, test conditions, and source)
- Source value
- 6.4
- Score
- 6.4
- Metric
- reported processing time for one 28.32-second source conversion
- Unit
- seconds
- Benchmark version
- Quantslant Voice Changer test; conversion run 2026-09-29
- Category
- audio
- Direction
- Lower is better
- Sample
- 1
- Source type
- independent-evaluator
- Evaluator
- Quantslant voice-changer test
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. One 28.32-second synthetic English monologue was converted through the API with eleven_multilingual_sts_v2 from voice Chris to library voice Lily. The result table reports elapsed processing time.
- Context
- One author-observed run, not a controlled latency benchmark; hardware, network path, concurrency, and timing boundaries are not stated. Quantslant discloses an ElevenLabs partner-link commission.
Quantslant clean conversion processing time, Chris to designed narrator (Show measurement, test conditions, and source)
- Source value
- 6
- Score
- 6
- Metric
- reported processing time for one 28.32-second source conversion
- Unit
- seconds
- Benchmark version
- Quantslant Voice Changer test; conversion run 2026-09-29
- Category
- audio
- Direction
- Lower is better
- Sample
- 1
- Source type
- independent-evaluator
- Evaluator
- Quantslant voice-changer test
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. One 28.32-second synthetic English monologue was converted through the API with eleven_multilingual_sts_v2 from voice Chris to a previously designed low English narrator voice. The result table reports elapsed processing time.
- Context
- One author-observed run, not a controlled latency benchmark; hardware, network path, concurrency, and timing boundaries are not stated. Quantslant discloses an ElevenLabs partner-link commission.
Quantslant noisy conversion content errors without background-noise removal (Show measurement, test conditions, and source)
- Source value
- 23
- Score
- 23
- Metric
- Scribe transcription errors against the 88-word English source after voice conversion
- Unit
- count
- Benchmark version
- Quantslant Voice Changer test; noisy conversion run 2026-09-29
- Category
- audio
- Direction
- Lower is better
- Sample
- 1
- Source type
- independent-evaluator
- Evaluator
- Quantslant voice-changer test
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. The same 88-word English source was mixed with cafe chatter at a volume equal to the speech, converted to library voice Lily with eleven_multilingual_sts_v2, and processed without remove_background_noise. Scribe transcription was compared with the source.
- Context
- One noisy recording and one conversion per condition; this is a combined model-plus-input-processing observation, not a general noise-robustness benchmark. Transcription uses Scribe, and Quantslant discloses an ElevenLabs partner-link commission.
Quantslant noisy conversion content errors with background-noise removal (Show measurement, test conditions, and source)
- Source value
- 2
- Score
- 2
- Metric
- Scribe transcription errors against the 88-word English source after voice conversion
- Unit
- count
- Benchmark version
- Quantslant Voice Changer test; noisy conversion run 2026-09-29
- Category
- audio
- Direction
- Lower is better
- Sample
- 1
- Source type
- independent-evaluator
- Evaluator
- Quantslant voice-changer test
- Status
- active
- Retrieved at
- 2026-10-15
- Methodology
- Independent evaluation. The same 88-word English source was mixed with cafe chatter at a volume equal to the speech, converted to library voice Lily with eleven_multilingual_sts_v2, and processed with remove_background_noise=true. Scribe transcription was compared with the source.
- Context
- One noisy recording and one conversion per condition; this is a combined model-plus-input-processing observation, not a general noise-robustness benchmark. Transcription uses Scribe, and Quantslant discloses an ElevenLabs partner-link commission.
Popularity
0 / 100
Sources and data date
| Type | Evidence and data date |
|---|---|
| Research status | Research date October 15, 2026. 6 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication. |
| Not found in the inspected sources | standardized-independent-quality-benchmark Note An exact-model independent editorial experiment was found, but no standardized leaderboard or replicated quality benchmark for eleven_multilingual_sts_v2 was identified. |
| Not found in the inspected sources | immutable-api-revision Note The experiment identifies the exact request model ID eleven_multilingual_sts_v2 but publishes no immutable provider checkpoint or revision. Results therefore attach to the API label used on the dated run, not a version-pinned artifact. |
| Additional source | ElevenLabs model documentation (retrieved September 24, 2026; October 15, 2026) · ElevenLabs Models documentation · Editorial description reviewed October 2, 2026 |
| Additional source | Quantslant voice-changer test (retrieved October 15, 2026) |
| Additional source | Model metadata (retrieved October 15, 2026) |
| Additional source | Model metadata (retrieved October 15, 2026) |
| Additional source | Model metadata (retrieved October 15, 2026) |
| Additional source | OpenRouter: Popularity index (derived), CC BY 4.0 |
| Additional source | Vercel AI Gateway: Popularity index (derived), CC BY 4.0 |
| Additional source | Creative Commons Attribution 4.0 |