NobleX-AI/vi-stt-mimi-vivoice

viVoice: Vietnamese speech as Mimi codes with timed transcripts

1,513.3 hours, 1,297,642 clips, 22,247,884 words, 186 channels of Vietnamese speech from viVoice (Thinh Le Phuoc Gia et al.) (186 Vietnamese YouTube channels: audiobooks, podcasts, news), stored as Kyutai Mimi codes with word-timed transcripts.

It is one source of the ~20,000 h on which NobleX-AI trains a streaming Vietnamese speech recognizer (Kyutai's delayed-streams STT, kyutai/stt-1b-en_fr, fine-tuned for Vietnamese). The other sources released so far: yodas2, vsv1100, fosd, bud500.

• No audio files. Each clip is its Mimi encoding (mimi-pytorch-e351c8d8@125.safetensors of kyutai/stt-1b-en_fr: 12.5 frames per second, 32 codebooks, int16), which is what the STT model reads. The codes decode back to 24 kHz audio (lossy, see Usage).
• The corpus' own transcripts, normalised to the spoken form in lower case (numbers, dates and units read out as words), with casing and punctuation added by the dragonSwing/xlm-roberta-capu tagger (CC-BY-SA-4.0). About 505 h of the clips are noise / reverberation augmented copies of other clips (the same channel and transcript).
• Timing. Every word has its start and end in seconds within its clip (forced alignment).

Files

file	content
transcripts.parquet	one row per clip: clip, channel, offset and n_frames (its rows in codes.npy), duration (s), text (cased, punctuated), words (lower case) with start / end (s)
codes.npy	int16 [total_frames, 32]: the Mimi codes of all clips back to back
clips.npz	per clip: offset, n_frames, dur, channel (index into meta.json channels), src
words.npz	per word: clip, onset, end (s), tok_off, n_tok (its ids in tokens.npy)
tokens.npy	int16 token ids of all words; itos.json maps an id to its string (▁x starts a word with letter x, ##x continues it, other entries are whole syllables)
labels/labels_0.npz	per word: case (0 as is, 1 Capitalised, 2 UPPER, 3 other: the string is in labels/mixed_0.json under the word's index) and punct (0 none, else 1 + index in , . ? : !, the mark after the word)
vocab.json, meta.json	token vocabulary (syllables); channel names (channels are @<YouTube channel handle>), hours, counts
MANIFEST.sha256	checksums of every file (sha256sum -c MANIFEST.sha256)

Usage

import numpy as np, pyarrow.parquet as pq
from huggingface_hub import snapshot_download

root = snapshot_download("NobleX-AI/vi-stt-mimi-vivoice", repo_type="dataset")
rows = pq.read_table(f"{root}/transcripts.parquet")
codes = np.load(f"{root}/codes.npy", mmap_mode="r")
r = rows.slice(0, 1).to_pylist()[0]
clip_codes = codes[r["offset"]:r["offset"] + r["n_frames"]]         # [T, 32] Mimi codes, 12.5 Hz
print(r["text"], list(zip(r["words"], r["start"], r["end"]))[:5])

# back to audio (pip install moshi)
import torch
from moshi.models import loaders
from huggingface_hub import hf_hub_download
mimi = loaders.get_mimi(hf_hub_download("kyutai/stt-1b-en_fr", "mimi-pytorch-e351c8d8@125.safetensors"), device="cpu")
mimi.set_num_codebooks(32)
with torch.no_grad():
    wav = mimi.decode(torch.from_numpy(clip_codes.astype(np.int64).T[None]))  # [1, 1, T * 1920] at 24 kHz

License

This dataset is a modified version of viVoice (Thinh Le Phuoc Gia, Tuan Pham Minh, Hung Nguyen Quoc, Trung Nguyen Quoc, Vinh Truong Hoang), licensed under CC-BY-NC-SA-4.0, and is released under the same license: non-commercial use only, share-alike. Changes: the audio was encoded as Mimi codes (no audio is distributed), part of it after noise / reverberation augmentation, and the transcripts were normalised and punctuated as described above. The upstream authors state that viVoice is intended for research purposes only, and that experiments conducted with it must adhere to ethical guidelines and legal standards, ensuring no harm to individuals and organizations; the same applies here.

Casing and punctuation were added with the dragonSwing/xlm-roberta-capu tagger (CC-BY-SA-4.0), credited here with thanks.

Citation

Please cite the upstream dataset:

@misc{viVoice,
  author = {Thinh Le Phuoc Gia, Tuan Pham Minh, Hung Nguyen Quoc, Trung Nguyen Quoc, Vinh Truong Hoang},
  title = {viVoice: Enabling Vietnamese Multi-Speaker Speech Synthesis},
  url = {https://github.com/thinhlpg/viVoice},
  year = {2024}
}