mirror of
https://github.com/modelscope/FunASR
synced 2025-09-15 14:48:36 +08:00
Merge pull request #435 from alibaba-damo-academy/dev_dzh
add docs for speaker verification and diarization
This commit is contained in:
zhifu gao
GitHub
commit
5ccd35ad3e
@ -1,20 +0,0 @@
|
||||
# Speaker Diarization
|
||||
|
||||
## Inference with pipeline
|
||||
|
||||
### Quick start
|
||||
|
||||
### Inference with you data
|
||||
|
||||
### Inference with multi-threads on CPU
|
||||
|
||||
### Inference with multi GPU
|
||||
|
||||
## Finetune with pipeline
|
||||
|
||||
### Quick start
|
||||
|
||||
### Finetune with your data
|
||||
|
||||
## Inference with your finetuned model
|
||||
|
||||
1
docs/modelscope_pipeline/sd_pipeline.md
Symbolic link
1
docs/modelscope_pipeline/sd_pipeline.md
Symbolic link
@ -0,0 +1 @@
|
||||
../../egs_modelscope/speaker_diarization/TEMPLATE/README.md
|
||||
@ -1,20 +0,0 @@
|
||||
# Speaker Verification
|
||||
|
||||
## Inference with pipeline
|
||||
|
||||
### Quick start
|
||||
|
||||
### Inference with you data
|
||||
|
||||
### Inference with multi-threads on CPU
|
||||
|
||||
### Inference with multi GPU
|
||||
|
||||
## Finetune with pipeline
|
||||
|
||||
### Quick start
|
||||
|
||||
### Finetune with your data
|
||||
|
||||
## Inference with your finetuned model
|
||||
|
||||
1
docs/modelscope_pipeline/sv_pipeline.md
Symbolic link
1
docs/modelscope_pipeline/sv_pipeline.md
Symbolic link
@ -0,0 +1 @@
|
||||
../../egs_modelscope/speaker_verification/TEMPLATE/README.md
|
||||
81
egs_modelscope/speaker_diarization/TEMPLATE/README.md
Normal file
81
egs_modelscope/speaker_diarization/TEMPLATE/README.md
Normal file
@ -0,0 +1,81 @@
|
||||
# Speaker Diarization
|
||||
|
||||
> **Note**:
|
||||
> The modelscope pipeline supports all the models in
|
||||
[model zoo](https://alibaba-damo-academy.github.io/FunASR/en/modelscope_models.html#pretrained-models-on-modelscope)
|
||||
to inference and finetine. Here we take the model of xvector_sv as example to demonstrate the usage.
|
||||
|
||||
## Inference with pipeline
|
||||
### Quick start
|
||||
```python
|
||||
from modelscope.pipelines import pipeline
|
||||
from modelscope.utils.constant import Tasks
|
||||
|
||||
# initialize pipeline
|
||||
inference_diar_pipline = pipeline(
|
||||
mode="sond_demo",
|
||||
num_workers=0,
|
||||
task=Tasks.speaker_diarization,
|
||||
diar_model_config="sond.yaml",
|
||||
model='damo/speech_diarization_sond-zh-cn-alimeeting-16k-n16k4-pytorch',
|
||||
reversion="v1.0.5",
|
||||
sv_model="damo/speech_xvector_sv-zh-cn-cnceleb-16k-spk3465-pytorch",
|
||||
sv_model_revision="v1.2.2",
|
||||
)
|
||||
|
||||
# input: a list of audio in which the first item is a speech recording to detect speakers,
|
||||
# and the following wav file are used to extract speaker embeddings.
|
||||
audio_list = [
|
||||
"https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_data/speaker_diarization/record.wav",
|
||||
"https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_data/speaker_diarization/spk1.wav",
|
||||
"https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_data/speaker_diarization/spk2.wav",
|
||||
"https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_data/speaker_diarization/spk3.wav",
|
||||
"https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_data/speaker_diarization/spk4.wav",
|
||||
]
|
||||
|
||||
results = inference_diar_pipline(audio_in=audio_list)
|
||||
print(results)
|
||||
```
|
||||
|
||||
#### API-reference
|
||||
##### Define pipeline
|
||||
- `task`: `Tasks.speaker_diarization`
|
||||
- `model`: model name in [model zoo](https://alibaba-damo-academy.github.io/FunASR/en/modelscope_models.html#pretrained-models-on-modelscope), or model path in local disk
|
||||
- `ngpu`: `1` (Default), decoding on GPU. If ngpu=0, decoding on CPU
|
||||
- `output_dir`: `None` (Default), the output path of results if set
|
||||
- `batch_size`: `1` (Default), batch size when decoding
|
||||
- `smooth_size`: `83` (Default), the window size to perform smoothing
|
||||
- `dur_threshold`: `10` (Default), segments shorter than 100 ms will be dropped
|
||||
- `out_format`: `vad` (Default), the output format, choices `["vad", "rttm"]`.
|
||||
- vad format: spk1: [1.0, 3.0], [5.0, 8.0]
|
||||
- rttm format: "SPEAKER test1 0 1.00 2.00 <NA> <NA> spk1 <NA> <NA>" and "SPEAKER test1 0 5.00 3.00 <NA> <NA> spk1 <NA> <NA>"
|
||||
|
||||
##### Infer pipeline for speaker embedding extraction
|
||||
- `audio_in`: the input to process, which could be:
|
||||
- list of url: `e.g.`: waveform files at a website
|
||||
- list of local file path: `e.g.`: path/to/a.wav
|
||||
- ("wav.scp,speech,sound", "profile.scp,profile,kaldi_ark"): a script file of waveform files and another script file of speaker profiles (extracted with the [model](https://www.modelscope.cn/models/damo/speech_xvector_sv-zh-cn-cnceleb-16k-spk3465-pytorch/summary))
|
||||
```text
|
||||
wav.scp
|
||||
test1 path/to/enroll1.wav
|
||||
test2 path/to/enroll2.wav
|
||||
|
||||
profile.scp
|
||||
test1 path/to/profile.ark:11
|
||||
test2 path/to/profile.ark:234
|
||||
```
|
||||
The profile.ark file contains speaker embeddings in a kaldi-like style.
|
||||
Please refer [README.md](../../speaker_verification/TEMPLATE/README.md) for more details.
|
||||
|
||||
### Inference with you data
|
||||
For single input, we recommend the "list of local file path" mode for inference.
|
||||
For multiple inputs, we recommend the last mode with pre-organized wav.scp and profile.scp.
|
||||
|
||||
### Inference with multi-threads on CPU
|
||||
We recommend the last mode with split wav.scp and profile.scp. Then, run inference for each split part.
|
||||
Please refer [README.md](../../speaker_verification/TEMPLATE/README.md) to find a similar process.
|
||||
|
||||
### Inference with multi GPU
|
||||
Similar to CPU, please set `ngpu=1` for inference on GPU.
|
||||
Besides, you should use `CUDA_VISIBLE_DEVICES=0` to specify a GPU device.
|
||||
Please refer [README.md](../../speaker_verification/TEMPLATE/README.md) to find a similar process.
|
||||
121
egs_modelscope/speaker_verification/TEMPLATE/README.md
Normal file
121
egs_modelscope/speaker_verification/TEMPLATE/README.md
Normal file
@ -0,0 +1,121 @@
|
||||
# Speaker Verification
|
||||
|
||||
> **Note**:
|
||||
> The modelscope pipeline supports all the models in
|
||||
[model zoo](https://alibaba-damo-academy.github.io/FunASR/en/modelscope_models.html#pretrained-models-on-modelscope)
|
||||
to inference and finetine. Here we take the model of xvector_sv as example to demonstrate the usage.
|
||||
|
||||
## Inference with pipeline
|
||||
|
||||
### Quick start
|
||||
#### Speaker verification
|
||||
```python
|
||||
from modelscope.pipelines import pipeline
|
||||
from modelscope.utils.constant import Tasks
|
||||
|
||||
inference_sv_pipline = pipeline(
|
||||
task=Tasks.speaker_verification,
|
||||
model='damo/speech_xvector_sv-zh-cn-cnceleb-16k-spk3465-pytorch'
|
||||
)
|
||||
|
||||
# The same speaker
|
||||
rec_result = inference_sv_pipline(audio_in=(
|
||||
'https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/sv_example_enroll.wav',
|
||||
'https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/sv_example_same.wav'))
|
||||
print("Similarity", rec_result["scores"])
|
||||
|
||||
# Different speakers
|
||||
rec_result = inference_sv_pipline(audio_in=(
|
||||
'https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/sv_example_enroll.wav',
|
||||
'https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/sv_example_different.wav'))
|
||||
print("Similarity", rec_result["scores"])
|
||||
```
|
||||
#### Speaker embedding extraction
|
||||
```python
|
||||
from modelscope.pipelines import pipeline
|
||||
from modelscope.utils.constant import Tasks
|
||||
|
||||
# Define extraction pipeline
|
||||
inference_sv_pipline = pipeline(
|
||||
task=Tasks.speaker_verification,
|
||||
model='damo/speech_xvector_sv-zh-cn-cnceleb-16k-spk3465-pytorch'
|
||||
)
|
||||
# Extract speaker embedding
|
||||
rec_result = inference_sv_pipline(
|
||||
audio_in='https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/sv_example_enroll.wav')
|
||||
speaker_embedding = rec_result["spk_embedding"]
|
||||
```
|
||||
Full code of demo, please ref to [infer.py](https://github.com/alibaba-damo-academy/FunASR/blob/main/egs_modelscope/speaker_verification/speech_xvector_sv-zh-cn-cnceleb-16k-spk3465-pytorch/infer.py).
|
||||
|
||||
#### API-reference
|
||||
##### Define pipeline
|
||||
- `task`: `Tasks.speaker_verification`
|
||||
- `model`: model name in [model zoo](https://alibaba-damo-academy.github.io/FunASR/en/modelscope_models.html#pretrained-models-on-modelscope), or model path in local disk
|
||||
- `ngpu`: `1` (Default), decoding on GPU. If ngpu=0, decoding on CPU
|
||||
- `output_dir`: `None` (Default), the output path of results if set
|
||||
- `batch_size`: `1` (Default), batch size when decoding
|
||||
- `sv_threshold`: `0.9465` (Default), the similarity threshold to determine
|
||||
whether utterances belong to the same speaker (it should be in (0, 1))
|
||||
|
||||
##### Infer pipeline for speaker embedding extraction
|
||||
- `audio_in`: the input to process, which could be:
|
||||
- url (str): `e.g.`: https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/sv_example_enroll.wav
|
||||
- local_path: `e.g.`: path/to/a.wav
|
||||
- wav.scp: `e.g.`: path/to/wav1.scp
|
||||
```text
|
||||
wav.scp
|
||||
test1 path/to/enroll1.wav
|
||||
test2 path/to/enroll2.wav
|
||||
```
|
||||
- bytes: `e.g.`: raw bytes data from a microphone
|
||||
- fbank1.scp,speech,kaldi_ark: `e.g.`: extracted 80-dimensional fbank features
|
||||
with kaldi toolkits.
|
||||
|
||||
##### Infer pipeline for speaker verification
|
||||
- `audio_in`: the input to process, which could be:
|
||||
- Tuple(url1, url2): `e.g.`: (https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/sv_example_enroll.wav, https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/sv_example_different.wav)
|
||||
- Tuple(local_path1, local_path2): `e.g.`: (path/to/a.wav, path/to/b.wav)
|
||||
- Tuple(wav1.scp, wav2.scp): `e.g.`: (path/to/wav1.scp, path/to/wav2.scp)
|
||||
```text
|
||||
wav1.scp
|
||||
test1 path/to/enroll1.wav
|
||||
test2 path/to/enroll2.wav
|
||||
|
||||
wav2.scp
|
||||
test1 path/to/same1.wav
|
||||
test2 path/to/diff2.wav
|
||||
```
|
||||
- Tuple(bytes, bytes): `e.g.`: raw bytes data from a microphone
|
||||
- Tuple("fbank1.scp,speech,kaldi_ark", "fbank2.scp,speech,kaldi_ark"): `e.g.`: extracted 80-dimensional fbank features
|
||||
with kaldi toolkits.
|
||||
|
||||
### Inference with you data
|
||||
Use wav1.scp or fbank.scp to organize your own data to extract speaker embeddings or perform speaker verification.
|
||||
In this case, the `output_dir` should be set to save all the embeddings or scores.
|
||||
|
||||
### Inference with multi-threads on CPU
|
||||
You can inference with multi-threads on CPU as follow steps:
|
||||
1. Set `ngpu=0` while defining the pipeline in `infer.py`.
|
||||
2. Split wav.scp to several files `e.g.: 4`
|
||||
```shell
|
||||
split -l $((`wc -l < wav.scp`/4+1)) --numeric-suffixes wav.scp splits/wav.scp.
|
||||
```
|
||||
3. Start to extract embeddings
|
||||
```shell
|
||||
for wav_scp in `ls splits/wav.scp.*`; do
|
||||
infer.py ${wav_scp} outputs/$((basename ${wav_scp}))
|
||||
done
|
||||
```
|
||||
4. The embeddings will be saved in `outputs/*`
|
||||
|
||||
### Inference with multi GPU
|
||||
Similar to inference on CPU, the difference are as follows:
|
||||
|
||||
Step 1. Set `ngpu=1` while defining the pipeline in `infer.py`.
|
||||
|
||||
Step 3. specify the gpu device with `CUDA_VISIBLE_DEVICES`:
|
||||
```shell
|
||||
for wav_scp in `ls splits/wav.scp.*`; do
|
||||
CUDA_VISIBLE_DEVICES=1 infer.py ${wav_scp} outputs/$((basename ${wav_scp}))
|
||||
done
|
||||
```
|
||||
15
egs_modelscope/speaker_verification/TEMPLATE/infer.py
Normal file
15
egs_modelscope/speaker_verification/TEMPLATE/infer.py
Normal file
@ -0,0 +1,15 @@
|
||||
from modelscope.pipelines import pipeline
|
||||
from modelscope.utils.constant import Tasks
|
||||
import sys
|
||||
|
||||
# Define extraction pipeline
|
||||
inference_sv_pipline = pipeline(
|
||||
task=Tasks.speaker_verification,
|
||||
model='damo/speech_xvector_sv-zh-cn-cnceleb-16k-spk3465-pytorch',
|
||||
output_dir=sys.argv[2],
|
||||
)
|
||||
# Extract speaker embedding
|
||||
rec_result = inference_sv_pipline(
|
||||
audio_in=sys.argv[1],
|
||||
|
||||
)
|
||||
Loading…
Reference in New Issue
Block a user