Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.
Python asraudiochineseemotion-recognitionfunasrmcp-servermultilingual-asropenai-compatible-apiparaformerpunctuationpytorchreal-time-asrspeaker-diarizationspeech-recognitionspeech-to-textstreaming-asrtranscriptionvllmvoice-activity-detectionwhisper-alternative
feat: add agent-friendly CLI (funasr audio.wav --output-format json)
游雁b8351fd361c0
diff --git a/README.md b/README.mdindex 626386c82..7f7981b91 100644--- a/README.md+++ b/README.md@@ -197,6 +197,31 @@ model = AutoModel(model="emotion2vec_plus_large", device="cuda")result = model.generate(input="audio.wav", granularity="utterance")```++### CLI (Agent-Friendly)++```bash+# Transcribe audio (simplest)+funasr audio.wav++# JSON output (for AI agents)+funasr audio.wav --output-format json++# SRT subtitles+funasr audio.wav --output-format srt --output-dir ./subs++# Speaker diarization + timestamps+funasr audio.wav --spk --timestamps -f json++# Choose model and language+funasr audio.wav --model paraformer --language zh++# Batch transcribe+funasr *.wav --output-format srt --output-dir ./output+```++Available models: `sensevoice` (default), `paraformer`, `paraformer-en`, `fun-asr-nano`+---## Deploydiff --git a/README_zh.md b/README_zh.mdindex 919d20fce..bab10b386 100644--- a/README_zh.md+++ b/README_zh.md@@ -195,6 +195,31 @@ model = AutoModel(model="emotion2vec_plus_large", device="cuda")result = model.generate(input="audio.wav", granularity="utterance")```+### 命令行工具(Agent 友好)++```bash+# 转写音频(最简用法)+funasr audio.wav++# JSON 输出(适合 AI Agent 调用)+funasr audio.wav --output-format json++# 生成 SRT 字幕+funasr audio.wav --output-format srt --output-dir ./subs++# 说话人分离 + 时间戳+funasr audio.wav --spk --timestamps -f json++# 指定模型和语言+funasr audio.wav --model paraformer --language zh++# 批量转写+funasr *.wav --output-format srt --output-dir ./output+```++可用模型:`sensevoice`(默认)、`paraformer`、`paraformer-en`、`fun-asr-nano`++---## 部署diff --git a/funasr/cli.py b/funasr/cli.pynew file mode 100644index 000000000..7107c0ea1--- /dev/null+++ b/funasr/cli.py@@ -0,0 +1,157 @@+"""FunASR CLI - Agent-friendly speech recognition from the command line."""++import argparse+import json+import os+import re+import sys+import time++MODEL_CONFIGS = {+ "sensevoice": {"model": "iic/SenseVoiceSmall", "vad_model": "fsmn-vad", "vad_kwargs": {"max_single_segment_time": 30000}},+ "paraformer": {"model": "paraformer-zh", "vad_model": "fsmn-vad", "punc_model": "ct-punc"},+ "paraformer-en": {"model": "paraformer-en", "vad_model": "fsmn-vad"},+ "fun-asr-nano": {"model": "FunAudioLLM/Fun-ASR-Nano"},+}+++def clean_text(text):+ return re.sub(r"<\|[^|]*\|>", "", text).strip()+++def _srt_time(ms):+ s = ms / 1000.0+ h, m, sec = int(s // 3600), int((s % 3600) // 60), int(s % 60)+ return f"{h:02d}:{m:02d}:{sec:02d},{int((s % 1) * 1000):03d}"+++def format_srt(segments):+ lines = []+ for i, seg in enumerate(segments, 1):+ lines += [str(i), f"{_srt_time(seg.get('start',0))} --> {_srt_time(seg.get('end',0))}", seg.get('text',''), ""]+ return "\n".join(lines)+++def format_tsv(segments):+ lines = ["start\tend\ttext"]+ for seg in segments:+ lines.append(f"{seg.get('start',0)/1000:.3f}\t{seg.get('end',0)/1000:.3f}\t{seg.get('text','')}")+ return "\n".join(lines)+++def _format_output(text, segments, timestamps, fmt, audio_path, model_name, language, elapsed):+ if fmt == "text":+ return text+ elif fmt == "json":+ obj = {"text": text}+ if segments:+ obj["segments"] = segments+ if timestamps:+ obj["timestamps"] = timestamps+ obj.update({"file": os.path.basename(audio_path), "model": model_name, "language": language or "auto", "duration_s": round(elapsed, 3)})+ return json.dumps(obj, ensure_ascii=False, indent=2)+ elif fmt == "srt":+ return format_srt(segments) if segments else f"1\n00:00:00,000 --> 99:59:59,999\n{text}\n"+ elif fmt == "tsv":+ return format_tsv(segments) if segments else f"start\tend\ttext\n0.000\t0.000\t{text}"+++def _get_version():+ try:+ from funasr import __version__+ return __version__+ except Exception:+ return "unknown"+++def main():+ p = argparse.ArgumentParser(+ prog="funasr",+ description="FunASR - speech recognition CLI. 50+ languages, speaker diarization.",+ epilog="Examples:\n"+ " funasr audio.wav\n"+ " funasr audio.wav --model sensevoice -f json\n"+ " funasr audio.wav -f srt -o ./subs\n"+ " funasr audio.wav --spk --timestamps\n",+ formatter_class=argparse.RawDescriptionHelpFormatter,+ )+ p.add_argument("audio", nargs="+", help="Audio file(s) to transcribe")+ p.add_argument("--model", "-m", default="sensevoice", choices=list(MODEL_CONFIGS), help="Model (default: sensevoice)")+ p.add_argument("--language", "-l", default=None, help="Language: zh, en, ja, ko, yue, auto")+ p.add_argument("--device", default=None, help="Device: cuda:0, cpu (default: auto)")+ p.add_argument("--output-format", "-f", default="text", choices=["text", "json", "srt", "tsv"], help="Output format (default: text)")+ p.add_argument("--output-dir", "-o", default=None, help="Write output files to directory")+ p.add_argument("--timestamps", action="store_true", help="Include word-level timestamps")+ p.add_argument("--spk", action="store_true", help="Enable speaker diarization")+ p.add_argument("--hotwords", default=None, help="Comma-separated hotwords")+ p.add_argument("--verbose", "-v", action="store_true", help="Show loading/timing info on stderr")+ p.add_argument("--version", action="version", version=f"%(prog)s {_get_version()}")+ args = p.parse_args()++ if args.verbose:+ print(f"Loading model: {args.model} ...", file=sys.stderr)++ import torch+ from funasr import AutoModel++ device = args.device or ("cuda:0" if torch.cuda.is_available() else "cpu")+ config = MODEL_CONFIGS[args.model].copy()+ if args.spk and "spk_model" not in config:+ config["spk_model"] = "cam++"+ if "punc_model" not in config and args.model not in ("fun-asr-nano", "sensevoice"):+ config["punc_model"] = "ct-punc"++ t_load = time.time()+ model = AutoModel(device=device, disable_update=True, **config)+ if args.verbose:+ print(f"Model loaded in {time.time() - t_load:.1f}s", file=sys.stderr)++ if args.output_dir:+ os.makedirs(args.output_dir, exist_ok=True)++ for audio_path in args.audio:+ if not os.path.isfile(audio_path):+ print(f"Error: file not found: {audio_path}", file=sys.stderr)+ sys.exit(1)++ if args.verbose:+ print(f"Transcribing: {audio_path}", file=sys.stderr)++ t0 = time.time()+ gen_kw = {"input": audio_path, "batch_size": 1}+ if args.language:+ gen_kw["language"] = args.language+ if args.hotwords:+ gen_kw["hotwords"] = args.hotwords.split(",")++ result = model.generate(**gen_kw)+ elapsed = time.time() - t0++ text = clean_text(result[0].get("text", ""))+ segments = []+ if "sentence_info" in result[0]:+ for seg in result[0]["sentence_info"]:+ s = {"start": seg.get("start", 0), "end": seg.get("end", 0), "text": clean_text(seg.get("text", ""))}+ if args.spk and "spk" in seg:+ s["speaker"] = seg["spk"]+ segments.append(s)++ timestamps = result[0].get("timestamps") if args.timestamps else None+ output = _format_output(text, segments, timestamps, args.output_format, audio_path, args.model, args.language, elapsed)++ if args.output_dir:+ ext = {"text": "txt", "json": "json", "srt": "srt", "tsv": "tsv"}[args.output_format]+ out_path = os.path.join(args.output_dir, os.path.splitext(os.path.basename(audio_path))[0] + "." + ext)+ with open(out_path, "w", encoding="utf-8") as f:+ f.write(output)+ if args.verbose:+ print(f"Written: {out_path}", file=sys.stderr)+ else:+ print(output)++ if args.verbose:+ print(f"Done in {elapsed:.2f}s", file=sys.stderr)+++if __name__ == "__main__":+ main()diff --git a/setup.py b/setup.pyindex 624ec0aa6..fdc693e1d 100644--- a/setup.py+++ b/setup.py@@ -152,7 +152,8 @@ setup(],entry_points={"console_scripts": [- "funasr = funasr.bin.inference:main_hydra",+ "funasr = funasr.cli:main",+ "funasr-hydra = funasr.bin.inference:main_hydra","funasr-server = funasr.bin.server:main","funasr-train = funasr.bin.train:main_hydra","funasr-train-ds = funasr.bin.train_ds:main_hydra",