somu9/vctk-tokens

MintTTS Pre-tokenized Audio Tokens

Pre-extracted audio codec tokens for TTS training.

Source

• Dataset: sanchit-gandhi/vctk
• Codec: MOSS-Audio-Tokenizer-Nano
• Codec sample rate: 48,000 Hz (stereo)
• Frame rate: 12.5 Hz (1 frame = 80ms)

Stats

Metric	Value
Total samples	88,156
Total audio hours	81.5h
Codebooks	16
Avg frames/sample	41.6
Avg duration	3.3s

Format

JSONL file (manifest.jsonl) where each line is:

{
  "text": "The transcribed text",
  "audio_codes": [[cb0, cb1, ..., cb15], ...],
  "n_frames": 125,
  "n_codebooks": 16
}

• audio_codes: List of [n_frames, 16] — each frame has 16 codebook tokens (0-1023)
• n_frames: Number of audio frames (duration = n_frames / 12.5 seconds)
• n_codebooks: Always 16

Usage

import json

with open("manifest.jsonl") as f:
    for line in f:
        sample = json.loads(line)
        text = sample["text"]
        codes = sample["audio_codes"]  # [n_frames, 16]
        print(f"{text[:50]}... | {len(codes)} frames ({len(codes)/12.5:.1f}s)")

Codec

Tokens were extracted using MOSS-Audio-Tokenizer-Nano (22M params). To decode tokens back to audio, use:

from transformers import AutoModel
import torch

codec = AutoModel.from_pretrained(
    "OpenMOSS-Team/MOSS-Audio-Tokenizer-Nano",
    trust_remote_code=True,
).eval()

codes = torch.tensor(sample["audio_codes"])  # [T, 16]
codes = codes.T.contiguous()                 # [16, T]
decoded = codec.batch_decode([codes], num_quantizers=16, chunk_duration=None)
waveform = decoded.audio[0]                  # [2, samples] at 48kHz