pdf-icon

Product Guide

Offline Voice Recognition

Industrial Control

IoT Measuring Instruments

Air Quality

Module13.2 PPS

Ethernet Camera

DIP Switch Usage Guide

Module GPS v2.0

Module GNSS

Module ExtPort For Core2

Module LoRa868 V1.2

Whisper

  • This section only guides you on how to run the precompiled voice-to-text example based on Whisper Small on Raspberry Pi 5;
  • For model conversion and example source code compilation, please refer to whisper.axcl.
  1. Download
git clone https://github.com/ml-inory/whisper.axcl.git
  1. Compile
cd whisper.axcl
mkdir -p build && cd build
cmake -DCMAKE_INSTALL_PREFIX=../install -DCMAKE_BUILD_TYPE=Release ..
make install -j4
  1. Precompiled Model

Download the precompiled model from Baidu Netdisk

  1. Run Whisper
cd install
./whisper -w ../demo.wav

Execution Result:

(base) axera@raspberrypi:~/qtang/whisper.axcl/install $ ./whisper -w ../demo.wav
encoder: ../models/small-encoder.axmodel
decoder_main: ../models/small-decoder-main.axmodel
decoder_loop: ../models/small-decoder-loop.axmodel
wav_file: ../demo.wav
language: zh
Load encoder take 3336.25 ms
Load decoder_main take 6091.89 ms
Load decoder_loop take 5690.05 ms
Read positional_embedding
Encoder run take 190.26 ms
First token: 17556       take 51.49ms
Next Token: 20844        take 30.15 ms
Next Token: 7781         take 30.21 ms
Next Token: 20204        take 30.20 ms
Next Token: 28455        take 30.17 ms
Next Token: 31962        take 30.02 ms
Next Token: 6336         take 30.09 ms
Next Token: 254          take 30.22 ms
Next Token: 2930         take 30.14 ms
Next Token: 236          take 30.14 ms
Next Token: 36135        take 30.12 ms
Next Token: 15868        take 30.18 ms
Next Token: 252          take 30.01 ms
Next Token: 1546         take 30.17 ms
Next Token: 46514        take 30.17 ms
Next Token: 50257        take 30.15 ms
All Token: take 503.68ms, 31.77 token/s
All take 735.09ms
Result: Even a near standstill in trading occurred.
(base) axera@raspberrypi:~/qtang/whisper.axcl/install $
On This Page