typhoon-asr-qwen-1.7b-ctx

Thai ASR with prompt-based contextual biasing: a LoRA fine-tune (merged, 2.07B params) of Qwen/Qwen3-ASR-1.7B-hf that accepts a list of names/terms in the system prompt, boosts the ones that are actually spoken, writes them with the listed spelling, and ignores the rest (distractor false alarm rate ≤.0004; non-speech probe 5/5). The strongest CER under biasing of every system measured below.

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-qwen-1.7b-ctx	2.07B	4.86	6.87
typhoon-asr-qwen-0.6b-ctx	0.80B	5.06†	7.28
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
typhoon-whisper-medium	769M	4.81	7.66
typhoon-asr-realtime-nemo-ctc	135M	8.54	13.46
typhoon-asr-realtime (RNN-T)	109M	6.89	9.92
nectec/Pathumma-whisper-th-large-v3	1.55B	5.84	10.36
biodatlab/whisper-th-large-v3-combined	1.55B	15.78†	14.91

† single-digit repetition-loop utterances dominate (0.6b: one utt; Thonburian: three, excluding them 8.39).

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); the two typhoon-qwen models are LoRA fine-tunes; typhoon-asr-realtime-nemo-ctc spots-and-splices at decode time.

	typhoon-asr-qwen-1.7b-ctx	typhoon-asr-qwen-0.6b-ctx	typhoon-asr-realtime-nemo-ctc	typhoon-whisper-medium (ctx)	typhoon-whisper-large-v3	nectec/Pathumma-whisper-th-large-v3	biodatlab/whisper-th-large-v3-combined	Qwen3-ASR-1.7B (original)
params	2.07B	0.80B	135M	769M	1.55B	1.55B	1.55B	2.07B
none (baseline)	5.11 / .441	5.71 / .397	9.56 / .158	7.05 / .383	4.83 / .459	5.23 / .416	5.68 / .482	5.25 / .396
bias@0 (gold)	3.39 / .754	4.01 / .668	5.97 / .852	4.92 / .658	N/A	3.40 / .823	6.37 / .842	4.48 / .644
bias@10	4.02 / .717	4.91 / .607	7.07 / .849	8.09 / .597	N/A	6.01 / .772	8.72 / .783	5.97 / .603
bias@50	4.30 / .640	5.25 / .530	7.51 / .839	N/A	N/A	N/A	N/A	6.71 / .522
bias@500	4.98 / .483	5.96 / .411	11.12 / .757	N/A	N/A	N/A	N/A	5.87 / .413
distractor false alarms ↓	≤.0004	≤.0004	.003–.006	.003	—	.0003	.0033	.0125

N/A: typhoon-whisper-large-v3's prompt pathway collapses on any list (echo loops, CER 200+); the whisper fine-tunes cannot take lists beyond ~10–15 entries because Whisper's 448-token decoder window is shared between the prompt and the transcript. Only the typhoon-qwen pair and the typhoon-asr-realtime-nemo-ctc spotter survive large lists.

Usage

import torch, soundfile as sf
from transformers import AutoModelForMultimodalLM, AutoProcessor  # transformers>=5.13

mid = "typhoon-ai/typhoon-asr-qwen-1.7b-ctx"
processor = AutoProcessor.from_pretrained(mid)
model = AutoModelForMultimodalLM.from_pretrained(mid, dtype=torch.bfloat16,
                                                 device_map="cuda").eval()
audio, sr = sf.read("clip.wav", dtype="float32")   # 16 kHz mono

def transcribe(system_text):
    msgs = [{"role": "system", "content": [{"type": "text", "text": system_text}]},
            {"role": "user",   "content": [{"type": "audio", "audio": audio}]}]
    inputs = processor.apply_chat_template([msgs], tokenize=True,
        add_generation_prompt=True, return_dict=True, return_tensors="pt").to(model.device)
    out = model.generate(**inputs, max_new_tokens=440)
    out = out[:, inputs["input_ids"].shape[1]:]
    return processor.decode(out, return_format="transcription_only")[0].strip()

print(transcribe("Thai"))                                   # plain ASR
print(transcribe("วโรรส Facebook ก้าวไกล\nThai"))            # with a bias list

Prompt format is strict: the system message is the space-joined bias list, then Thai on its own line — or just Thai with no list. Recall decays gracefully with list length; ≤50 entries is the sweet spot.

License & attribution

Apache-2.0 — LoRA-merged weights of Qwen/Qwen3-ASR-1.7B-hf (Apache-2.0). Trained on ~10,800 hours of Thai speech (GigaSpeech2-th); see the Typhoon ASR Real-time technical report.