Voice Lab
The Voice Lab is where you create and manage cloned voice profiles for use with Fish-Speech engines. Record or upload a short audio clip, and KoKoFish builds a reusable voice that appears in the voice dropdown throughout the app.
Kokoro users: The Voice Lab is disabled when Kokoro is the active engine. Kokoro uses built-in preset voices — see the Speech Lab for voice selection and blending.
- Go to the Voice Lab tab
- Click Upload Audio or Record to provide a reference clip
- Give the voice a name
- Click Create Profile — KoKoFish encodes the reference audio into a voice embedding
- The new voice appears in the voice dropdown in the Speech Lab
Reference audio guidelines:
- Minimum: ~3 seconds of clear speech
- Ideal: 15–60 seconds of natural, varied speech
- Maximum: 180 seconds — longer clips are trimmed automatically
- Use clean audio with minimal background noise for best results
- A single continuous monologue works better than a conversation with two speakers
Profiles are stored per engine so your 1.4 and 1.5 voice libraries stay separate:
voices/
fish14/ ← Fish-Speech 1.4 voice profiles
fish15/ ← Fish-Speech 1.5 voice profiles
Switching engines shows only the voices for that engine.
- Rename — change the display name of a profile
- Delete — permanently removes the profile
- Preview — play a test audio clip using the selected voice
- Voice cloning quality improves significantly with a longer, cleaner reference clip. A 30-second clip recorded in a quiet room will outperform a 5-second noisy clip every time.
- Fish-Speech 1.5 tends to produce more faithful voice clones than 1.4 — if voice accuracy is important, 1.5 is worth the extra download.
- You can create multiple profiles from the same person at different recording qualities or emotional states and pick the best one per use.
- Fish-Speech voice clones can speak other languages if you feed them translated text — the voice characteristics (timbre, rhythm, pitch) carry over even if the language changes.