#I want to migrate my v2 voices to v3. How do I go about doing that?

1 messages · Page 1 of 1 (latest)

glad dirgeBOT
#
AI-generated Answer for harry018485

Short answer: there’s no “migration” step. Keep using the same voice_id(s) and switch the model to v3 when you generate audio.

How to use your existing voices with v3

  • In the web app: select the Eleven v3 model from the model dropdown and generate as usual with your saved voices. (help.elevenlabs.io)
  • Via API (Text to Speech): call POST /v1/text-to-speech/:voice_id and set model_id to eleven_v3 (your existing voice_id stays the same). (elevenlabs.io)
  • Via API (Text to Dialogue): your inputs include the same voice_id values; model_id defaults to eleven_v3 unless you override it. (elevenlabs.io)
  • Availability: Eleven v3 is generally available via the API as of August 20, 2025. (elevenlabs.io)

Notes for Professional Voice Clones (PVCs)

  • PVCs are fine‑tuned per model; by default they train on Flash v2.5, Turbo v2.5 and Multilingual v2. When we add fine‑tuning support for new models, you’ll see a notification in My Voices and can start the additional fine‑tune from there. (help.elevenlabs.io)
  • If you try to use a PVC before its fine‑tuning is ready, you may see “No model found for this voice.” Wait for training to complete in My Voices. (help.elevenlabs.io)

Optional: creating new voices for v3

  • Voices made with Voice Design v3 are fully compatible with Eleven v3 and support its expressive audio tags; they’re also backward compatible with other models. (help.elevenlabs.io)

If you tell me whether you’re using the web app or API (and whether your voices are PVC or IVC), I can give exact steps or example requests.

Sources:
How do I get access to Eleven v3 (Alpha)? – ElevenLabs
Create speech | ElevenLabs Documentation
Create dialogue | ElevenLabs Documentation
August 20, 2025 | ElevenLabs Documentation
How do I add or upgrade the models used to train my Professional Voice Clone? – ElevenLabs
What does the error 'No model found for this voice. Please select another voice' mean? – ElevenLabs
What is Voice Design? – ElevenLabs

trail bone
#

But don't newer voices sound better with v3? My old v2 voices won't support emotion tags, will they?

glad dirgeBOT
#
AI-generated Answer for harry018485

Short answer:

  • Yes—v3 is designed to produce more human‑like, higher‑quality, and more expressive speech than earlier models. It’s in public alpha and not ideal for real‑time use. (help.elevenlabs.io)
  • Emotion/audio tags are a v3 model feature. If you synthesize your existing voices with modelId=eleven_v3, you can use tags like [whispers], [sad], [laughs]; effectiveness varies by voice and its training data. (help.elevenlabs.io)
  • Important caveat: Professional Voice Clones created earlier aren’t yet fully optimized for v3 and may respond less reliably to tags; ElevenLabs suggests using Instant Voice Clones or Voice Design v3 voices for best v3 tag performance until PVC optimization lands. (elevenlabs.io)

How to try it:

  • Call Text-to-Speech with your existing voice_id and set modelId to eleven_v3; include inline tags (e.g., “[whispers] … [sighs] …”). (elevenlabs.io)

If you need consistent, real‑time behavior, stay on Turbo/Flash for now and use v3 where the extra expressiveness matters. (help.elevenlabs.io)

Sources:
What is Eleven v3 (Alpha)? – ElevenLabs
What is Eleven v3 (Alpha)? – ElevenLabs
Eleven v3: Most Expressive AI Text to Speech Model Launched | ElevenLabs
Eleven v3 (alpha) — The most expressive Text to Speech model
What is Eleven v3 (Alpha)? – ElevenLabs

trail bone
#

Is there any way to normalize the speed of all voices to be the same speed?

glad dirgeBOT
#
AI-generated Answer for harry018485

Short answer: you can set the same speed value for every voice, but there’s no global “normalize all voices” switch. Set the speed per request or update each voice’s settings to the same value.

Details:

  • Speed control is supported across Text to Speech (web and API), Studio, and Agents. Valid range: 0.7–1.2, default 1.0. Values <1.0 slow down; >1.0 speed up. (help.elevenlabs.io)
  • API: each voice has a speed field in its settings. You can read and set it via:
    • GET /v1/voices/:voice_id/settings (returns speed) (elevenlabs.io)
    • POST /v1/voices/:voice_id/settings/edit (set speed to your target across all voices in a loop) (elevenlabs.io)
    • You can also just pass the speed parameter per synthesis request. (elevenlabs.io)
  • Important: even with identical speed values, pacing can still differ slightly by voice because cadence is influenced by the original training/audio samples. Speed helps, but it’s not a perfect cross-voice normalizer. (elevenlabs.io)

If you want a uniform target, pick a value (e.g., 1.0) and set it for all voices via the edit settings endpoint, or include the same speed in every TTS call. (elevenlabs.io)

Sources:
Can I change the pace of the voice? – ElevenLabs
Get voice settings | ElevenLabs Documentation
Edit voice settings | ElevenLabs Documentation
Text to Speech (product guide) | ElevenLabs Documentation
Controls | ElevenLabs Documentation
Edit voice settings | ElevenLabs Documentation

trail bone
#

Can I confirm that all voices with the same dialogue will always return a voice file with the same duration if their speed parameter is set to the same value?

glad dirgeBOT
#
AI-generated Answer for harry018485

Short answer: No.

  • ElevenLabs TTS is non-deterministic, so repeated generations with the same text/parameters can vary. Even when a seed is provided, determinism is “not guaranteed.” (elevenlabs.io)
  • Pace depends on the specific voice (how it was created/trained). The speed control is just a multiplier (0.7–1.2) and does not equalize timing across different voices. (elevenlabs.io)
  • The Text-to-Speech “Create speech” API has no parameter to force a target duration; only speed (via voice settings) and an optional seed are available for this purpose. (elevenlabs.io)

If you need tighter timing control:

  • Insert explicit pauses with <break time="x.xs" /> for predictable gaps. (help.elevenlabs.io)
  • Measure resulting word/phrase timings with the Forced Alignment API. (elevenlabs.io)

Sources:
Text to Speech (product guide) | ElevenLabs Documentation
Controls | ElevenLabs Documentation
Create speech | ElevenLabs Documentation
Do pauses and SSML phoneme tags work with the API? – ElevenLabs
Forced Alignment | ElevenLabs Documentation

trail bone
#

Is the seed parameter still supported by v3 in the Text to Speech API?