speech_recognition/FunASR
1
0
Fork 0
mirror of https://github.com/modelscope/FunASR synced 2025-09-15 14:48:36 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

funasr1.0

Browse Source
This commit is contained in:
游雁 2023-12-21 21:28:14 +08:00
parent f2a68d0bc8
commit b66a41fb73
1 changed files with 37 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

56
README_zh.md
View File

@ -92,30 +92,48 @@ print(res)
```
注：`model_hub`：表示模型仓库，`ms`为选择modelscope下载，`hf`为选择huggingface下载。
### 实时语音识别
```python
from funasr import infer
[//]: # (### 实时语音识别)
p = infer(model="paraformer-zh-streaming", model_hub="ms")
[//]: # (```python)
chunk_size = [0, 10, 5] #[0, 10, 5] 600ms, [0, 8, 4] 480ms
param_dict = {"cache": dict(), "is_final": False, "chunk_size": chunk_size, "encoder_chunk_look_back": 4, "decoder_chunk_look_back": 1}
[//]: # (from funasr import infer)
import torchaudio
speech = torchaudio.load("asr_example_zh.wav")[0][0]
speech_length = speech.shape[0]
[//]: # ()
[//]: # (p = infer(model="paraformer-zh-streaming", model_hub="ms"))
stride_size = chunk_size[1] * 960
sample_offset = 0
for sample_offset in range(0, speech_length, min(stride_size, speech_length - sample_offset)):
param_dict["is_final"] = True if sample_offset + stride_size >= speech_length - 1 else False
input = speech[sample_offset: sample_offset + stride_size]
rec_result = p(input=input, param_dict=param_dict)
print(rec_result)
```
注：`chunk_size`为流式延时配置，`[0,10,5]`表示上屏实时出字粒度为`10*60=600ms`，未来信息为`5*60=300ms`。每次推理输入为`600ms`（采样点数为`16000*0.6=960`），输出为对应文字，最后一个语音片段输入需要设置`is_final=True`来强制输出最后一个字。
[//]: # ()
[//]: # (chunk_size = [0, 10, 5] #[0, 10, 5] 600ms, [0, 8, 4] 480ms)
更多详细用法（[新人文档](https://alibaba-damo-academy.github.io/FunASR/en/funasr/quick_start_zh.html)）
[//]: # (param_dict = {"cache": dict(), "is_final": False, "chunk_size": chunk_size, "encoder_chunk_look_back": 4, "decoder_chunk_look_back": 1})
[//]: # ()
[//]: # (import torchaudio)
[//]: # (speech = torchaudio.load("asr_example_zh.wav")[0][0])
[//]: # (speech_length = speech.shape[0])
[//]: # ()
[//]: # (stride_size = chunk_size[1] * 960)
[//]: # (sample_offset = 0)
[//]: # (for sample_offset in range(0, speech_length, min(stride_size, speech_length - sample_offset)):)
[//]: # ( param_dict["is_final"] = True if sample_offset + stride_size >= speech_length - 1 else False)
[//]: # ( input = speech[sample_offset: sample_offset + stride_size])
[//]: # ( rec_result = p(input=input, param_dict=param_dict))
[//]: # ( print(rec_result))
[//]: # (```)
[//]: # (注：`chunk_size`为流式延时配置，`[0,10,5]`表示上屏实时出字粒度为`10*60=600ms`，未来信息为`5*60=300ms`。每次推理输入为`600ms`（采样点数为`16000*0.6=960`），输出为对应文字，最后一个语音片段输入需要设置`is_final=True`来强制输出最后一个字。)
[//]: # ()
[//]: # (更多详细用法（[新人文档](https://alibaba-damo-academy.github.io/FunASR/en/funasr/quick_start_zh.html)）)
<a name="服务部署"></a>