fish-audio/voice-changer
Fish Audio speech-to-speech voice changer. Keeps the words, timing and delivery of your source recording but speaks them in a voice cloned from a second audio clip.
Model Inputs
Source speech to convert (.mp3/.wav/.flac/.m4a/.ogg/.opus/.aac/.webm/.mp4). Its words, timing and delivery are kept; only the voice is replaced
Drag & drop a file here, or click to select
Supported formats: MP3, WAV, FLAC, M4A, OGG, OPUS, AAC, WEBM, MP4 (max 100MB)
Audio of the target voice (.mp3/.wav/.flac/.m4a/.ogg/.opus/.aac/.webm/.mp4). It is cloned, used for the conversion, then discarded. 30 seconds or more of clean single-speaker speech works best
Drag & drop a file here, or click to select
Supported formats: MP3, WAV, FLAC, M4A, OGG, OPUS, AAC, WEBM, MP4 (max 100MB)
Result
Pricing
fish-audio/
voice-changer
Pricing for Synexa AI models works differently from other providers. Instead of being billed by time, you are billed by input and output, making pricing more predictable.
For example, generating 100 requests costs about $100.00.
Check out our docs for more information about how per-request pricing works on Synexa.
| Provider | Price ($) | Saving (%) |
|---|---|---|
| Synexa | $1.00 | - |
| replicate | $2.00 | 50.0% |
Readme
Fish Audio speech-to-speech turns one recording into another voice. You give it two clips: audio carries the performance — the words, the pacing, the emotion — and reference_audio carries the voice you want to hear. The output says exactly what the source said, in the target voice.
Capabilities:
- Voice conversion that preserves the source delivery, not just the transcript — pauses, emphasis and emotion survive the swap
- Target voice is cloned instantly from a single reference clip; no training step and no persistent voice library
- Output is MP3
Usage notes:
- Both clips accept .mp3, .wav, .flac, .m4a, .ogg, .opus, .webm, .mp4 and .aac
- Clean, single-speaker audio gives the best result on both sides; music or overlapping voices in the reference degrade the clone
- 30 seconds or more of reference audio is recommended; very short samples produce a looser match
- The pitch and timbre follow the reference, so cross-gender conversions shift the voice substantially
- Cost scales with the duration of the source clip, not with the reference clip
Limitations:
- Any container outside the list above is rejected before the request is sent
- A source clip with no speech in it, or one that exceeds the account allowance, fails the conversion rather than returning silent audio