Skip to content

FunASR Chinese Speech Recognition:

FunASR is an open-source speech recognition model from Alibaba. It performs better than the whisper series for Chinese speech scenarios. It's supported via HTTP calls in video translation software using the zh_recogn and SenseVoice projects. You only need to deploy the corresponding zh_recogn and SenseVoice integration packages. After starting them, simply fill in the API address in the video translation software to use it.

However, many users still find this operation confusing. Therefore, starting from v2.97, this functionality has been integrated into the video translation software. You no longer need to separately deploy and start the zh_recogn and SenseVoice projects. Simply select FunASR Chinese Speech Recognition in the software to use it.

image.png

Selecting FunASR Chinese in Speech Recognition

After selecting FunASR Chinese Speech Recognition in speech recognition, you can choose to use the paraformer-zh model or the SenseVoiceSmall model. The former is recommended, offering better performance and speed than the latter.

image.png

Downloading the FunASR Model Online for the First Use

To avoid excessively large software packages, the FunASR model is not integrated into the software package. It will be automatically downloaded from modelscope.cn the first time you use it. After downloading, it's saved to the models folder under the software directory, within the hub folder. Depending on your network conditions, this may take a few minutes to several tens of minutes. Just patiently wait for the download to complete as long as there are no red errors.

image.png

Downloading the FunASR Model File Separately in Advance

If your download speed is slow and you want to download the model file in advance, you can click this link: https://pan.baidu.com/s/1HNvH_riUADtnZew8QjYS9w?pwd=jt5r, download it, and unzip it. You will see a hub folder. Copy this folder to the models folder in the software directory, overwriting the existing hub folder.

As shown, after downloading, there is a hub folder in the compressed package, and there is also a hub folder in the models directory under the software directory.

image.png