Audio Generator

Script

0 characters

Voice settings

Channel

Checkpoint

Reference voice

Generation

Speed1.00x

Quality (NFE steps)32

Advanced

Guidance strength (CFG)2.0

Cross-fade duration0.15s

Target loudness (RMS)0.10

Remove silence

Seed

Studio mastering

Apply mastering chain

Highpass filter65 Hz

Compression3.2:1

Gain boost+2.2 dB

Noise gate

Create or select project

A project holds one channel's training data, transcripts, vocab, and checkpoints.

Project name

Tokenizer type

Existing projects

village_healer_charactive

Audio files

Files uploaded to this project's raw data folder.

2.mp3 ... 26.mp322 / 25 uploaded

Transcription result

Not run yet.

Run

Language

Manual review pass

Vocab check result

Not run yet.

Actions

Extend only if the check reports missing symbols (e.g. for a new language).

Dataset stats

691

samples

4h08m

audio

2545

vocab size

1.58s

shortest clip

Actions

Chinese tokenizer

Live status - not running -

14,800

step

21/100

epoch

57%

GPU util

0.31

loss

Waiting for connection...

Hyperparameters

Model

Learning rate

Batch size / GPU

Batch type

Epochs

Warmup updates

Save every

Keep last N

Fine-tune (vs train from scratch)

Mixed precision

fp16 confirmed stable on this GPU - bf16 crashes the vocoder during sample logging.

Reference

Reference text

Reference audio

reference_voice.wav

Generate

Text to generate

No output yet.

Checkpoint

Settings

Speed1.00x

NFE steps32

Jobs (0)

No jobs yet - submit one from the Generate tab.