Automatic Speech Recognition - ASR

This example runs a speech recognition model based on the BPU inference engine to automatically transcribe .wav audio files into corresponding text. The example code is located in the /app/cdev_demo/bpu/07_speech_sample/01_asr/ directory.

Model Description

• Introduction:

  The ASR (Automatic Speech Recognition) model converts audio signals into text. The input is a single-channel audio waveform (after sample rate conversion and normalization), and the output is a character-level token sequence. When used together with a vocabulary (vocab) file, it supports Chinese speech transcription. This example uses a quantized .hbm model.

• HBM Model Name: asr.hbm

• Input Format: Audio waveform, single-channel, sampled at 16kHz, with a maximum length of 30,000 samples.

• Output: Probability distribution (logits) over character tokens; after argmax decoding, mapped to recognized text.

Functionality Overview

• Model Loading

  Loads the ASR model and automatically parses its input/output shapes and quantization information.

• Input Preprocessing

  Reads audio using SoundFile (supports .wav) and performs the following steps:

  • Converts to single-channel
  • Resamples to target sample rate (default: 16kHz)
  • Normalizes to zero-mean and unit-variance (z-score)
  • Pads or truncates to a fixed length (e.g., 30,000 samples)
  • Supports generator-based processing for long audio, enabling streaming recognition.

• Inference Execution

  Performs inference using the .infer() method.

• Post-processing

  Extracts token indices from output logits and maps them to characters using the vocab dictionary file (in JSON format), producing the final transcribed text.

Environment Dependencies

Before compiling and running, ensure the following dependencies are installed:

sudo apt update
sudo apt install -y libgflags-dev libsndfile1-dev libsamplerate0-dev

Directory Structure

.
|-- CMakeLists.txt                  # CMake build script: target/dependency/include/link configuration
|-- README.md                       # Usage instructions (this file)
|-- inc
|   |-- asr.hpp                     # ASR inference wrapper header (interfaces for loading/preprocessing/inference/post-processing)
|   `-- audio_chunk_reader.hpp      # Audio chunk reader: reads file → resamples → outputs chunks
`-- src
    |-- asr.cc                      # ASR inference implementation: input writing, forward computation, CTC decoding, etc.
    |-- audio_chunk_reader.cc       # Chunk reading implementation: libsndfile + libsamplerate for streaming chunking
    `-- main.cc                     # Program entry point: argument parsing → loop over chunks → inference → concatenate transcribed text

Build Instructions

• Configuration and Compilation

  mkdir build && cd build
  cmake ..
  make -j$(nproc)

Model Download

If the model is not found during runtime, download it using the following command:

wget https://archive.d-robotics.cc/downloads/rdk_model_zoo/rdk_s100/asr/asr.hbm

Parameter Description

Parameter	Description	Default Value
--model_path	Path to the model file (.hbm)	/opt/hobot/model/s100/basic/asr.hbm
--test_sound	Path to the input audio file (.wav)	/app/res/assets/chi_sound.wav
--vocab_file	Vocabulary file (JSON), mapping class id → token	/app/res/labels/vocab.json

Quick Start

• Run the Model

  • Ensure you are in the build directory.
  • Run with default parameters:

    ./asr

  • Run with specified parameters:

    ./asr \
        --model_path /opt/hobot/model/s100/basic/asr.hbm \
        --test_sound /app/res/assets/chi_sound.wav \
        --vocab_file /app/res/labels/vocab.json

• View Results

  Upon successful execution, the result will be printed:

  I am Qwen, a large-scale language model developed by Alibaba Cloud.||

Notes

• For more information about deployment options or model support, please refer to the official documentation or contact platform technical support.