• English

      • Performance Optimization

      Thread Tuning

      # Auto-detect CPU cores (Recommended)
manyspeech --threads 0 asr -t offline -f audio.wav

# Manual set (e.g., 8 cores)
manyspeech --threads 8 asr -t offline -f audio.wav

      Model Precision

      PrecisionSpeedAccuracyMemory
      int8FastHighLow
      fp32SlowHighestHigh
      # Use int8 (Default, Recommended)
manyspeech asr -t offline --accuracy int8 -f audio.wav

# Use fp32 (Higher Accuracy)
manyspeech asr -t offline --accuracy fp32 -f audio.wav

      Processing Method

      MethodMemoryLatencyUse Case
      oneHighHigh (Wait for end)Short audio (<30s)
      chunkLowLow (Streaming)Long audio, Real-time
      # Chunk recommended for long audio
manyspeech asr -t offline -m chunk -f long_audio.wav

      VAD Optimization

      Use Silero-VAD in noisy environments:

      manyspeech --vad silero-vad-v6-onnx asr -t online -i mic

      Testing Tips

      • Use --debug for timing details.
      • Increase --threads to find optimal value (usually = CPU cores).
      • chunk saves more memory than one for long audio.