I'm trying to create a middle-aged female voice, but it seems like the results are either too young

[marsh tapir]

AI-generated Answer for prncsgateau

Here are the supported ways to target a “middle‑aged female” voice in ElevenLabs:

• Use Voice Design with an explicit age/style prompt and higher guidance scale. Include age and role (example from our docs: “middle‑aged female newsreader”). If the voice skews young/old, raise the Guidance scale so the generation follows your prompt more closely, and regenerate to compare the three options produced each time. You’re only charged for preview text characters. (help.elevenlabs.io)

• If you prefer to pick an existing voice, search the Voice Library and filter by Gender: Female and Age: Middle‑aged, then add suitable voices to My Voices. (help.elevenlabs.io)

• To keep the delivery you want while changing timbre, run your own performance through Voice Changer (speech‑to‑speech). It preserves tone and delivery while converting to the target voice. (help.elevenlabs.io)

• In Studio, you can fine‑tune output using the per‑voice settings (e.g., stability, similarity, speaker boost) to further customize the sound after you’ve chosen a voice. (help.elevenlabs.io)

• If precision and consistency are critical, consider a Professional Voice Clone (PVC) that matches the target age—PVCs offer the highest quality and consistency compared to Voice Design. (help.elevenlabs.io)

If you want, share a short clip of the target style; I can suggest a Voice Design prompt and guidance scale starting point based on it.

Sources:
What kind of voices can I create using Voice Design? – ElevenLabs
What is the Voice Library? – ElevenLabs
What is Speech-to-speech? – ElevenLabs
How can I change the voice and settings across multiple paragraphs in Studio? – ElevenLabs
What is Voice Design? – ElevenLabs

[dusk crag]

I've tried, but even so the voice either skews too young or too old, and also there are way too many artifacts like echos or room noise, or it sounds like static.

[marsh tapir]

AI-generated Answer for prncsgateau

Thanks for sticking with it—here’s how to get a clean, middle‑aged female voice without the “too young/too old” drift or audible artifacts:

• Pick the right model: v3 (Alpha) is the most expressive but less reliable; for fewer artifacts and steadier delivery use Turbo v2.5 or Flash v2.5. If you do use v3, expect to do more prompt tuning. (help.elevenlabs.io)

• Prompt Voice Design precisely and constrain it: include age and recording quality in the prompt (e.g., “middle‑aged female, neutral American accent, calm narrator, studio/‘perfect’ audio quality, no reverb”). If the voice still skews young/old, increase the Guidance scale to make it follow your description more strictly. (elevenlabs.io)

• TTS voice settings that reduce artifacts:

  • Stability: raise toward the “robust” side for steadier, less variable delivery. (elevenlabs.io)
  • Similarity: if you hear hiss/room tone carried over from a clone, lower Similarity so the model doesn’t mimic imperfections in the reference. (elevenlabs.io)
  • Style exaggeration: keep at 0; higher values can destabilize output. (elevenlabs.io)
  • Speed: keep near 1.0; extreme values can degrade quality. (elevenlabs.io)

• Generate to WAV to avoid compression artifacts. In TTS/Voice Changer, download WAV from your History; WAV/44.1kHz/16‑bit is supported. (help.elevenlabs.io)

• If you’re producing long‑form in Studio, use Export so Auto‑Regenerate can automatically re‑generate lines where it detects artifacts. (help.elevenlabs.io)

• If you’re cloning or using Voice Changer, artifacts usually come from noisy source audio. Use clean, single‑speaker recordings (IVC: 1–2 min of high‑quality audio) and, if needed, run them through Voice Isolator first. (help.elevenlabs.io)

• As a fast alternative, try a prebuilt voice filtered by Gender = Female and Age = Middle‑aged in the Voice Library. (help.elevenlabs.io)

If you want, paste your current prompt/settings and model choice and I’ll suggest exact tweaks grounded in the docs. (elevenlabs.io)

[dusk crag]

But when I go to voice design it only has the V3 model available now. I don't see options for V2. I'm very confused.

[marsh tapir]

AI-generated Answer for prncsgateau

Short answer: that’s expected. Voice Design in the dashboard now runs on the Eleven v3 voice‑design model, so you won’t see a V2 option inside Voice Design. Voices you create there are backward‑compatible, and you can render them on Multilingual v2 (or Flash/Turbo) when you synthesize. (elevenlabs.io)

What to do:

• Create and save the voice in Voice Design (v3). (elevenlabs.io)
• When generating audio, switch the synthesis model to “Multilingual v2” (or another v2 family model) in the UI’s model picker/Project Settings, then render your saved voice there. Studio’s default is Multilingual v2 and you can choose newer models (including v3) in Project Settings. (elevenlabs.io)
• Via API, use the appropriate model ID (e.g., eleven_multilingual_v2, eleven_flash_v2_5, eleven_turbo_v2_5). The official model list is here. (help.elevenlabs.io)

If you’re aiming to minimize artifacts and get a stable “middle‑aged” delivery, prefer Multilingual v2 for final synthesis; it’s the recommended high‑fidelity/stable model, whereas v3 (alpha) is more expressive but can require more prompt tuning. (elevenlabs.io)

If anything above doesn’t match what you see in the UI, tell me exactly where you’re generating (Speech Synthesis vs Studio), and I’ll map the clicks.

Sources:
Voices | ElevenLabs Documentation
Voice design | ElevenLabs Documentation
Studio overview | ElevenLabs Documentation
How do I find the model ID? – ElevenLabs
Models | ElevenLabs Documentation

[dusk crag]

Well that won't work with Voice Design now will it? Why are you giving me instructions for style exageration and etc when that's not available in voice design anymore? I'm trying to make a French voice but it seems like French especially tends to skew too young/old.

[marsh tapir]

AI-generated Answer for prncsgateau

✅ Search complete, generating response...

[dawn tusk]

@dusk crag 1. Voice Selection
Blend two voices for the best result:
Voice A: Neutral Female (soft, fresh tone)
Voice B: Narration Female (medium, mature tone)

Goal: Voice A adds clarity, Voice B adds depth and maturity.

2. Voice Blend
   Mix ratio: 40% Voice A + 60% Voice B
   This ratio gives a balanced middle-aged sound — not too young, not too old.

3. Pitch & Formant Adjustments
   Pitch: -0.5 to 0 (slightly lower if voice sounds too young)
   Formant Shift: +0.1 to +0.2 (adds depth and maturity)

4. Style / Emotion
   Choose Warm / Conversational / Calm
   Avoid “Theatrical” or “Excited” styles; they alter perceived age.

5. Text Sample
   Use short, natural sentences
   Avoid very youthful slang or overly dramatic phrases

6. Final Tweaks
   Voice feels slightly too light → increase Formant +0.05
   Voice feels too heavy → slightly lower Pitch -0.2

[marsh tapir]

AI-generated Answer for tamara0n0

Short answer: ElevenLabs doesn’t support “voice blending” (mixing two voices in a single stream) or pitch/formant controls. Here’s what is supported and what to use instead, with links to the official docs.

• Voice selection and “blending”

  • There is no documented feature to blend two voices by ratio in TTS or Voice Design. If you need two voices in one passage, Studio lets you assign different voices to selected spans of text within the same paragraph (sequentially, not mixed). (help.elevenlabs.io)

• Style, emotion and delivery with Eleven v3

  • Eleven v3 (Alpha) adds audio tags for emotions, delivery, and human reactions (e.g., [sad], [angry], [whispers], [laughs], [sighs]). Use these to steer perceived delivery instead of “style exaggeration” sliders. (help.elevenlabs.io)
  • Voice Design v3 generates voices from descriptive prompts (e.g., accent, gender, pacing, tone, speaking style) and is meant to pair with Eleven v3. (help.elevenlabs.io)

• Pacing (speed)

  • You can control speaking speed via the Speed setting (0.7–1.2) across the web app and API. There are no pitch or formant parameters documented. (help.elevenlabs.io)

• French voices (accent/age realism)

  • Default and Voice Design voices are English; when used in other languages they may carry a slight English accent. For native French results, use a Professional Voice Clone in French (your own PVC) or pick a PVC from the Voice Library filtered by language/accent. (help.elevenlabs.io)
  • The Voice Library lets you filter by language/accent, gender, and age to find suitable PVCs. (help.elevenlabs.io)

• Reducing artifacts (echo/noise/static, glitches)

  • Use Studio’s Auto‑Regenerate when exporting; it automatically re-generates clips if mispronunciations or unwanted artefacts are detected. (help.elevenlabs.io)
  • If you encounter whispering/distortion, increase Stability (and regenerate) or try a different voice/model; these issues are often stability/voice dependent. (help.elevenlabs.io)
  • Avoid non-textual characters like { } < > [ ] in your script; they can cause low‑quality speech. (help.elevenlabs.io)

• Current model availability

  • Eleven v3 (Alpha) is available to all users and is the recommended pairing for Voice Design v3 voices; it supports audio tags and dialogue mode. (help.elevenlabs.io)

If you want, I can point you to

[dusk crag]

How do I blend voices? And do all of this other stuff because I don't see any of these options in the voice design too.