Create a configuration

README.md

@@ -12,11 +12,27 @@ python3 -m venv .venv
 ## Usage
 
 ```bash
-python stt_runner.py \
+python stt_runner.py [--language=Indonesian] audio1.wav audio2.wav ...
-    --conv-frontend=path/conv_frontend.onnx \
-    --encoder=path/encoder.onnx \
-    --decoder=path/decoder.onnx \
-    --tokenizer=path/tokenizer \
-    audio1.wav audio2.wav ...
 ```
 
+## Configuration
+
+Model paths and inference parameters are hardcoded in `config/`:
+
+- `config/model.py` — model paths (conv_frontend, encoder, decoder, tokenizer under `models/`)
+- `config/asr.py` — inference params: `LANGUAGE`, `HOTWORDS`, `NUM_THREADS`, `PROVIDER`, `SAMPLE_RATE`, `FEATURE_DIM`, `MAX_TOTAL_LEN`, `MAX_NEW_TOKENS`
+
+`LANGUAGE` defaults to `""` (all languages / auto-detect). Passing `--language` on the CLI overrides it.
+
+## Download Model (Qwen3-ASR 1.7B int8)
+
+```bash
+BASE="https://modelscope.cn/models/zengshuishui/Qwen3-ASR-onnx/resolve/master"
+mkdir -p models/model_1.7B models/tokenizer
+wget -O models/model_1.7B/conv_frontend.onnx "$BASE/model_1.7B/conv_frontend.onnx"
+wget -O models/model_1.7B/encoder.int8.onnx  "$BASE/model_1.7B/encoder.int8.onnx"
+wget -O models/model_1.7B/decoder.int8.onnx  "$BASE/model_1.7B/decoder.int8.onnx"
+for f in vocab.json merges.txt tokenizer_config.json preprocessor_config.json config.json chat_template.json; do
+  wget -O "models/tokenizer/$f" "$BASE/tokenizer/$f"
+done
+```

config/asr.py

@@ -0,0 +1,12 @@
+NUM_THREADS     = 2
+SAMPLE_RATE     = 16000
+FEATURE_DIM     = 128
+PROVIDER        = "cpu"
+MAX_TOTAL_LEN   = 2048
+MAX_NEW_TOKENS  = 256
+
+LANGUAGE        = "" # "" = support all languages / auto-detect
+HOTWORDS        = "" # Comma-separated hotword phrases, e.g. "AcmeCorp, FooBar".
+                     # Biases the model to transcribe these phrases correctly,
+                     # useful for brand/product/person names the model may
+                     # otherwise mishear. Leave empty to disable.

config/model.py

@@ -0,0 +1,8 @@
+from pathlib import Path
+
+MODEL_DIR = Path(__file__).resolve().parent.parent / "models"
+
+CONV_FRONTEND = MODEL_DIR / "model_1.7B" / "conv_frontend.onnx"
+ENCODER       = MODEL_DIR / "model_1.7B" / "encoder.int8.onnx"
+DECODER       = MODEL_DIR / "model_1.7B" / "decoder.int8.onnx"
+TOKENIZER     = MODEL_DIR / "tokenizer"

core/args_parser.py

@@ -1,14 +0,0 @@
-import argparse
-
-parser = argparse.ArgumentParser()
-parser.add_argument("--conv-frontend", type=str, required=True)
-parser.add_argument("--encoder", type=str, required=True)
-parser.add_argument("--decoder", type=str, required=True)
-parser.add_argument("--tokenizer", type=str, required=True)
-parser.add_argument("--language", type=str, default="", help="Force language, e.g. Indonesian, English, Chinese")
-parser.add_argument("--hotwords", type=str, default="", help="Comma-separated hotword phrases, e.g. 'foo,bar'")
-parser.add_argument("--num-threads", type=int, default=2)
-parser.add_argument("--provider", type=str, default="cpu", choices=["cpu", "cuda"])
-parser.add_argument("--max-total-len", type=int, default=2048)
-parser.add_argument("--max-new-tokens", type=int, default=256)
-parser.add_argument("sounds", nargs="+", help="Audio files to transcribe")

core/argsroom.py

@@ -0,0 +1,5 @@
+import argparse
+
+parser = argparse.ArgumentParser()
+parser.add_argument("--language", type=str, default=None, help="Force language, overrides config LANGUAGE (e.g. Indonesian, English, Chinese)")
+parser.add_argument("sounds", nargs="+", help="Audio files to transcribe")

stt_runner.py

@@ -1,23 +1,25 @@
 import sys
 from pathlib import Path
 import sherpa_onnx, soundfile as sf
-from core import args_parser as ap
+from core import argsroom as ap
+from config.model import CONV_FRONTEND, ENCODER, DECODER, TOKENIZER
+from config.asr import LANGUAGE, HOTWORDS, NUM_THREADS, SAMPLE_RATE, FEATURE_DIM, PROVIDER, MAX_TOTAL_LEN, MAX_NEW_TOKENS
 
 def stt_run(args):
 
     print("Recognize...")
     recognizer = sherpa_onnx.OfflineRecognizer.from_qwen3_asr( # qwen3 asr
-        conv_frontend   = args.conv_frontend,
+        conv_frontend   = str(CONV_FRONTEND),
-        encoder         = args.encoder,
+        encoder         = str(ENCODER),
-        decoder         = args.decoder,
+        decoder         = str(DECODER),
-        tokenizer       = args.tokenizer,
+        tokenizer       = str(TOKENIZER),
-        hotwords        = args.hotwords,
+        hotwords        = HOTWORDS,
-        num_threads     = args.num_threads,
+        num_threads     = NUM_THREADS,
-        sample_rate     = 16000,
+        sample_rate     = SAMPLE_RATE,
-        feature_dim     = 128,
+        feature_dim     = FEATURE_DIM,
-        provider        = args.provider,
+        provider        = PROVIDER,
-        max_total_len   = args.max_total_len,
+        max_total_len   = MAX_TOTAL_LEN,
-        max_new_tokens  = args.max_new_tokens,
+        max_new_tokens  = MAX_NEW_TOKENS,
     )
     print("Recognizer ready!")
 
@@ -31,8 +33,9 @@ def stt_run(args):
 
         stream     = recognizer.create_stream()
         
-        if args.language:
+        language   = args.language if args.language is not None else LANGUAGE
-            stream.set_option("language", args.language)
+        if language:
+            stream.set_option("language", language)
         stream.accept_waveform(sr, audio)
 
         recognizer.decode_stream(stream) # Inference execution for `stream.result`