moonshine-ai/moonshine-streaming-tiny-vi

Moonshine Streaming Tiny — Vietnamese

Vietnamese streaming speech recognition, 27.0M parameters. Same architecture as moonshine-ai/moonshine-streaming-tiny, trained for Vietnamese with a 12,288-entry Vietnamese tokenizer.

Moonshine Streaming pairs a 50 Hz time-domain audio frontend with a sliding-window Transformer encoder, so it transcribes incrementally rather than waiting for an utterance to finish. It is intended for on-device use on edge-class hardware.

Checkpoint identity

This repository is a conversion of one specific training checkpoint, recorded here because the weights behind a language move as later stages win:

Checkpoint	vi12k_stageC_best_macro9.395.safetensors
Stage	C (read-speech mix)
Architecture	slinkier_prime_adapted
Tokenizer	tokenizer_vi12k.json, vocab 12,288
Snapshot taken	2026-08-24
Parameters	27.0M

If you need reproducibility, pin the revision of this repository rather than tracking main.

Usage

pip install --upgrade transformers datasets[audio]

from transformers import MoonshineStreamingForConditionalGeneration, AutoProcessor
import torch

model = MoonshineStreamingForConditionalGeneration.from_pretrained(
    "moonshine-ai/moonshine-streaming-tiny-vi"
).eval()
processor = AutoProcessor.from_pretrained("moonshine-ai/moonshine-streaming-tiny-vi")

inputs = processor(audio, return_tensors="pt", sampling_rate=16000)

# Cap the output length. Like other seq2seq ASR models this one can fall into a
# repetition loop, and short or noisy clips are where it happens.
seq_lens = inputs.attention_mask.sum(dim=-1)
max_new_tokens = int((seq_lens * 6.5 / 16000).max().item()) + 2

generated = model.generate(**inputs, max_new_tokens=max_new_tokens)
print(processor.batch_decode(generated, skip_special_tokens=True)[0])

Pass the attention_mask. The encoder applies its per-layer sliding windows only when it is given one; called without a mask it attends over the whole utterance instead, which is a different model from the one that was trained. The processor returns the mask, so the snippet above is the safe form. The processor also pads audio to a whole number of 80-sample frames, which the frontend requires.

Architecture

Encoder	6 layers, width 320, 8 heads, sliding windows (16, 4) on the first two and last two layers and (16, 0) between
Decoder	6 layers, width 320, 8 heads, RoPE over 32 of each head's 40 dimensions
Frontend	50 Hz features, CMVN, asinh compression, two causal stride-2 convolutions
Adapter	learned absolute positional embeddings before the decoder

The lookahead layers give roughly 80 ms of lookahead; the intermediate layers have none.

Training data

Trained on a large-scale automatically labeled Vietnamese corpus, plus a much smaller read-speech set:

• Crawled corpus, roughly 83,000 hours, pseudo-labeled and unaudited.
• Track A read speech, roughly 700 hours.

The crawled transcripts are pseudo-labels: they were produced by running a Whisper-family teacher model over crawled audio, not by human transcription. The model therefore inherits the teacher's error modes, including its handling of proper nouns, numerals and code-switching. No human-verified transcript was used for the bulk of training.

Evaluation

Vietnamese is scored on word error rate (WER), after the usual case and punctuation normalization. Mandarin and Japanese in this model family are instead scored on no-space CER, because they are written without spaces; every other language, this one included, uses WER.

suite_vi is FLEURS Vietnamese (read speech) and LSVSC (a Vietnamese spontaneous-speech corpus), so unlike most of the languages in this family it is measured on both.

Seeded 400-utterance sample, batch 1

Batch 1 is the honest number for deployment. Batched evaluation zero-pads short clips up to the longest in the batch, and that trailing silence flatters the model.

Panel	WER
fleurs_vi	10.98
lsvsc_vi	7.63
macro	9.305

This repository against the training checkpoint

These weights were converted from the neo training checkpoint, and the conversion was checked by measurement rather than inspection: same seeded sample, same batch size, same normalizer. A conversion that loads and emits plausible text can still have a permuted weight mapping, which only a score catches.

	fleurs_vi	lsvsc_vi	macro
Training checkpoint	10.98	7.63	9.305
This repository	10.98	7.62	9.300

400/400 and 397/400 transcripts are byte-identical.

The quantized build we ship

The .ort package served to the Moonshine deployment library is quantized to int8 from these same weights, and scores 9.425 against 9.305 for the float checkpoint on the same sample under the same stopping rule -- a difference of +0.120, which is inside the noise of a 400-clip sample and should not be read as the quantized build being better or worse. That build is a different artifact from this repository, which is float32.

Limitations

• Machine-labeled training data. See above; the model reproduces its teacher's mistakes as well as its strengths.
• Repetition loops on short clips. Like other seq2seq ASR models this one can fall into a repetition loop, and short or noisy clips are where it happens. Cap the output length, as the usage snippet does.
• Evaluated on 2 panels only. No evaluation of telephony, children's speech, heavy dialect, or noisy far-field conditions.

Out-of-scope use

Not intended for non-consensual surveillance, speaker identification, or high-stakes decisions.

License

MIT.