Skip to content

Voices

Jump to bottom
aivrar edited this page Sep 25, 2026 · 1 revision

Home · Manual index

Voices tab

The Voices tab manages reference audio files stored in E:\tts_server\voices. These are not Kokoro's built-in af_heart names and not Edge's en-US-JennyNeural catalog. They are WAV/MP3/FLAC/OGG clips you clone from, condition on, or transcribe.

You can also drop files into E:\tts_server\voices in Explorer and click Refresh.

Layout

Voices tab showing the uploaded synthetic demo WAV and its playback and transcription controls

Cropped after the demo row to keep unrelated voice files private. This synthetic clip illustrates file management; the Kokoro take used the built-in af_heart voice.

  1. Upload Voice row: file picker (accept="audio/*", multiple) and primary Upload.
  2. Toolbar: Refresh, Select All, Deselect All, Delete Selected (N).
  3. Table of voices.

Upload

  1. Click the file input, choose one or more audio files (server cap 50 MB each).
  2. Click Upload.
  3. Each file is POST /api/voices/upload as multipart field file with the API token header.
  4. Toasts report success or failure per file; the table refreshes.

CLI equivalent:

tts.cmd voices upload C:\path\to\my-voice.wav

Keep clips clean, dry, single-speaker. Several engines document a sweet spot of 5–15 seconds. Long LibriVox excerpts reduced Fish Speech intelligibility in the guarded probe. OuteTTS cloning wants a clip no longer than 15 seconds plus its exact transcript.

Table columns

Column Content
Checkbox Multi-select for bulk delete
Name Display name (filename stem)
Format wav / mp3 / flac / ogg
Size B / KB / MB
Transcription Cached sidecar text, or -- if none
Actions Play, Stop, Transcribe, Delete

Empty state: No voice files found. Upload .wav/.mp3/.flac/.ogg files above, or drop them into the voices/ directory.

Play and Stop

Play fetches GET /api/voices/{filename}/audio with the token header, builds a blob URL, and plays it. The token never appears in the audio element's src. Stop calls App.stopAudio().

Transcribe

Transcribe runs Whisper on that file: POST /api/voices/{filename}/transcribe. Query defaults match the API: size=base, word_timestamps=false, optional language, task=transcribe (or translate). The GUI uses the defaults. The plain transcript is cached as <filename>.txt beside the audio (for bill_boerst_male.wav that is bill_boerst_male.wav.txt). The table then shows that text. Replacing the audio clears its transcript. Uploads are validated and published atomically; a transcription racing with replacement is rejected for retry. Legacy stem-only transcripts are read only when the audio filename is unambiguous.

You need Whisper installed (Setup card) and enough VRAM/CPU for the chosen size. The API will spawn a Whisper worker if needed.

CLI:

tts.cmd voices transcribe bill_boerst_male.wav --size base
tts.cmd voices info bill_boerst_male.wav

info prints size, sample rate, channels, frames, duration, format, transcription.

Delete

Row Delete removes that file and its .txt sidecar (DELETE /api/voices/{filename}).

Delete Selected (N) bulk-deletes checked rows (POST /api/voices/delete with {filenames: [...]}).

Select All / Deselect All toggle checkboxes. The selected set is pruned when a file disappears after Refresh.

CLI:

tts.cmd voices delete bill_boerst_male.wav
tts.cmd voices rename old.wav new.wav
tts.cmd voices download bill_boerst_male.wav --out C:\out\me.wav
tts.cmd voices list

Rename is not on the GUI; use CLI or POST /api/voices/{filename}/rename with {new_name} (1–128 characters). The sidecar is renamed with the audio.

Using a saved voice on the Testing tab

On Testing, engines with ref_audio: true show Saved Voice and Or Upload. Saved Voice is this same library. Picking a saved voice fills voice with the filename. Engines that also have ref_text: true show Reference Text — paste the exact transcript (the Voices-tab transcription is the usual source).

VibeVoice requires a reference (the Testing tab refuses Generate without one). XTTS can instead use a built-in named speaker with mode=built-in.

Bundled examples

The current installed copy has sample clips in voices\ (for example angie_female.wav, f5_english_ref.wav, lj_female.wav, british_male_p226.wav). Treat them as ready-to-clone references, not as Kokoro voice ids. The GitHub source checkout excludes the voices/ folder.

Related pages

Wiki home · Manual index · Download release · Source

Portable TTS Server V2 · Windows + WSL2 · Offline-ready Kokoro

Start here

Desktop tabs

Engines and audio

CLI and API

Maintenance

Reference

Download release

Clone this wiki locally