typhoon-asr-realtime-nemo-ctc

A Thai CTC model built entirely from standard NeMo classes built on typhoon-ai/typhoon-asr-realtime: EncDecCTCModelBPE whose ConformerEncoder carries the frozen 17 RT layers plus 4 newly trained layers (26M trainable) and a ConvASRDecoder over the RT BPE-2048 tokenizer. Loads with restore_from anywhere NeMo is installed; using only standard NeMo modules keeps the usual NeMo/Riva/TensorRT export paths.

Results

Plain transcription (CER%, same protocol: gigaspeech2 bench in-domain / TVSpeech out-of-domain 30-s talk-show):

model	params	gigaspeech2 CER ↓	TVSpeech CER ↓
typhoon-asr-realtime-nemo-ctc (greedy CTC)	135M	8.54	13.46
typhoon-asr-realtime (RNN-T)	109M	6.89	9.92
typhoon-whisper-medium	769M	4.81	7.66
typhoon-asr-qwen-1.7b-ctx	2.07B	4.86	6.87
Qwen3-ASR-1.7B (original)	2.07B	6.09	10.63
Qwen3-ASR-0.6B (original)	0.80B	8.20	11.45
typhoon-whisper-large-v3	1.55B	4.69	6.32
nectec/Pathumma-whisper-th-large-v3	1.55B	5.84	10.36
biodatlab/whisper-th-large-v3-combined	1.55B	15.78†	14.91

† dominated by 3 repetition-loop utterances; excluding them: 8.39.

Read this model's CER in role: it trades a few CER points against the larger autoregressive decoders for frame-level CTC posteriors — the thing that enables keyword spotting, word boosting, and timestamps below — while adding only 26M parameters on top of the typhoon-asr-realtime encoder.

Trained on ~10,800 hours of Thai speech (GigaSpeech2-th); see the Typhoon ASR Real-time technical report.

Contextual biasing — vs every route we measured

Same benchmark (wayu-ai/thai-contextasr-bench), same frozen scorer. Cells = CER% ↓ / entity recall ↑. Whisper models take the list as a decoder prompt (zero-shot); typhoon-asr-qwen-1.7b-ctx is a LoRA fine-tuned for biasing; this model spots and splices at decode time (word_boost.py, τ=−2).

	typhoon-asr-realtime-nemo-ctc	typhoon-asr-qwen-1.7b-ctx	typhoon-whisper-medium	typhoon-whisper-large-v3	nectec/Pathumma-whisper-th-large-v3	biodatlab/whisper-th-large-v3-combined	Qwen3-ASR-1.7B (original)
params	135M	2.07B	769M	1.55B	1.55B	1.55B	2.07B
none (baseline)	9.56 / .158	5.11 / .441	7.05 / .383	4.83 / .459	5.23 / .416	5.68 / .482	5.25 / .396
bias@0 (gold)	5.97 / .852	3.39 / .754	4.92 / .658	N/A	3.40 / .823	6.37 / .842	4.48 / .644
bias@10	7.07 / .849	4.02 / .717	8.09 / .597	N/A	6.01 / .772	8.72 / .783	5.97 / .603
bias@50	7.51 / .839	4.30 / .640	N/A	N/A	N/A	N/A	6.71 / .522
bias@500	11.12 / .757	4.98 / .483	N/A	N/A	N/A	N/A	5.87 / .413
distractor false alarms ↓	.003–.006	≤.0004	.003	—	.0003	.0033	.0125

N/A: typhoon-whisper's prompt pathway collapses on any list (echo loops, CER 200+ — erased by its prompt-free fine-tune); the fine-tuned medium and Pathumma/Thonburian cannot take lists beyond ~10–15 entries because Whisper's 448-token decoder window is shared between the prompt and the transcript. This model and typhoon-asr-qwen-1.7b-ctx have no list-size limit; only they remain standing past bias@10 — and this model's recall stays flat (.85 → .76 from 0 to 500 distractors) where every other route either dies or decays.

Rule of thumb: pick this model when entity recall, list-size robustness, compute, or streaming matter — it finds more of your keywords at ~1/15 the size with zero biasing training. Pick typhoon-asr-qwen-1.7b-ctx when absolute CER and tolerance to huge dirty lists matter most. They also compose: typhoon-qwen writes the transcript, this model supplies keyword hits and timestamps from one cheap encoder pass.

Usage

from nemo.collections.asr.models import EncDecCTCModelBPE
m = EncDecCTCModelBPE.restore_from("typhoon-asr-realtime-nemo-ctc.nemo")
print(m.transcribe(["clip.wav"])[0].text)

Keyword spotting on the CTC posteriors (bundled kws_spot.py):

python kws_spot.py --model typhoon-asr-realtime-nemo-ctc.nemo \
    --audio clip.wav --keywords "วโรรส,เฟซบุ๊ก"

Word boosting (bundled word_boost.py — spot then splice into the transcript, threshold --tau):

python word_boost.py --model typhoon-asr-realtime-nemo-ctc.nemo \
    --audio clip.wav --keywords "วโรรส,เฟซบุ๊ก" --tau -1.5

On thai-contextasr-bench (gold list): CER 9.56 → 5.97 with entity recall .158 → .852 at tau −2 (FA .003); tau −1.5 balanced (recall .758, FA .0014). Keep lists curated (≤50 entries) at aggressive tau.

Reference

Typhoon ASR: see the Typhoon ASR Real-time technical report.

License & attribution

CC-BY-4.0 — this checkpoint embeds encoder weights of typhoon-ai/typhoon-asr-realtime (CC-BY-4.0) verbatim; attribution to the Typhoon team is required. Newly trained layers and bundled scripts follow the same license.