feat: establish voice assistant V1 baseline

2026-10-09 18:31:21 +00:00
commit cebcb3fd61
57 changed files with 9501 additions and 0 deletions
--- a/tools/stt_smoke_test.py
+++ b/tools/stt_smoke_test.py
@@ -0,0 +1,114 @@
+import argparse
+import asyncio
+from pathlib import Path
+
+from riva.client.asr import AudioChunkFileIterator
+from riva.client.realtime import RealtimeClientASR
+
+
+NIM_SERVER = "127.0.0.1:9000"
+INPUT_FILE = "/home/ubuntu/voice-assistant/audios/chafik_normalize.wav"
+CHUNK_FRAMES = 1600
+
+
+def build_args() -> argparse.Namespace:
+    return argparse.Namespace(
+        input_file=INPUT_FILE,
+        mic=False,
+        list_devices=False,
+        duration=None,
+        input_device=None,
+
+        sample_rate_hz=16000,
+        num_channels=1,
+        file_streaming_chunk=CHUNK_FRAMES,
+
+        output_text=None,
+        prompt="",
+
+        server=NIM_SERVER,
+        ssl_root_cert=None,
+        ssl_client_cert=None,
+        ssl_client_key=None,
+        use_ssl=False,
+        metadata=None,
+        options=None,
+        max_message_length=67108864,
+
+        word_time_offsets=False,
+        max_alternatives=1,
+        profanity_filter=False,
+        automatic_punctuation=True,
+        no_verbatim_transcripts=False,
+
+        language_code="pt-BR",
+        model_name="",
+
+        boosted_lm_words=None,
+        boosted_lm_score=4.0,
+
+        speaker_diarization=False,
+        diarization_max_speakers=3,
+
+        start_history=-1,
+        start_threshold=-1.0,
+        stop_history=-1,
+        stop_threshold=-1.0,
+        stop_history_eou=-1,
+        stop_threshold_eou=-1.0,
+
+        custom_configuration="",
+        endpoint="/v1/realtime",
+        query_params="intent=transcription",
+        mint_api_key=None,
+    )
+
+
+async def main() -> None:
+    input_path = Path(INPUT_FILE)
+
+    if not input_path.exists():
+        raise FileNotFoundError(
+            f"Audio file not found: {input_path}"
+        )
+
+    args = build_args()
+
+    print(f"NIM server : {args.server}")
+    print(f"Audio      : {input_path}")
+    print(f"Language   : {args.language_code}")
+    print(f"Chunk      : {CHUNK_FRAMES} frames")
+
+    client = RealtimeClientASR(args=args)
+
+    audio_iterator = AudioChunkFileIterator(
+        input_file=str(input_path),
+        chunk_n_frames=CHUNK_FRAMES,
+        delay_callback=None,
+    )
+
+    try:
+        print("Connecting to NIM...")
+        await client.connect()
+        print("Connected.")
+
+        send_task = asyncio.create_task(
+            client.send_audio_chunks(audio_iterator)
+        )
+
+        receive_task = asyncio.create_task(
+            client.receive_responses()
+        )
+
+        await asyncio.gather(
+            send_task,
+            receive_task,
+        )
+
+    finally:
+        await client.disconnect()
+        print("Disconnected.")
+
+
+if __name__ == "__main__":
+    asyncio.run(main())