| | |
| | | ) |
| | | return model |
| | | ``` |
| | | This function defines the detail of the model. For different speech recognition models, the same speech recognition `Task` can usually be shared and the remaining thing needed to be done is to define a specific model in this function. For example, a speech recognition model with a standard encoder-decoder structure has been shown above. Specifically, it first defines each module of the model, including encoder, decoder, etc. and then combine these modules together to generate a complete model. In FunASR, the model needs to inherit `AbsESPnetModel` and the corresponding code can be seen in `funasr/train/abs_espnet_model.py`. The main function needed to be implemented is the `forward` function. |
| | | This function defines the detail of the model. For different speech recognition models, the same speech recognition `Task` can usually be shared and the remaining thing needed to be done is to define a specific model in this function. For example, a speech recognition model with a standard encoder-decoder structure has been shown above. Specifically, it first defines each module of the model, including encoder, decoder, etc. and then combine these modules together to generate a complete model. In FunASR, the model needs to inherit `FunASRModel` and the corresponding code can be seen in `funasr/train/abs_espnet_model.py`. The main function needed to be implemented is the `forward` function. |
| | | |
| | | Next, we take `SANMEncoder` as an example to introduce how to use a custom encoder as a part of the model when defining the specified model and the corresponding code can be seen in `funasr/models/encoder/sanm_encoder.py`. For a custom encoder, in addition to inheriting the common encoder class `AbsEncoder`, it is also necessary to define the `forward` function to achieve the forward computation of the `encoder`. After defining the `encoder`, it should also be registered in the `Task`. The corresponding code example can be seen as below: |
| | | ```python |
| | |
| | | ) |
| | | return model |
| | | ``` |
| | | 该函数定义了具体的模型。对于不同的语音识别模型,往往可以共用同一个语音识别`Task`,额外需要做的是在此函数中定义特定的模型。例如,这里给出的是一个标准的encoder-decoder结构的语音识别模型。具体地,先定义该模型的各个模块,包括encoder,decoder等,然后在将这些模块组合在一起得到一个完整的模型。在FunASR中,模型需要继承`AbsESPnetModel`,其具体代码见`funasr/train/abs_espnet_model.py`,主要需要实现的是`forward`函数。 |
| | | 该函数定义了具体的模型。对于不同的语音识别模型,往往可以共用同一个语音识别`Task`,额外需要做的是在此函数中定义特定的模型。例如,这里给出的是一个标准的encoder-decoder结构的语音识别模型。具体地,先定义该模型的各个模块,包括encoder,decoder等,然后在将这些模块组合在一起得到一个完整的模型。在FunASR中,模型需要继承`FunASRModel`,其具体代码见`funasr/train/abs_espnet_model.py`,主要需要实现的是`forward`函数。 |
| | | |
| | | 下面我们将以`SANMEncoder`为例,介绍如何在定义模型的时候,使用自定义的`encoder`来作为模型的组成部分,其具体的代码见`funasr/models/encoder/sanm_encoder.py`。对于自定义的`encoder`,除了需要继承通用的`encoder`类`AbsEncoder`外,还需要自定义`forward`函数,实现`encoder`的前向计算。在定义完`encoder`后,还需要在`Task`中对其进行注册,下面给出了相应的代码示例: |
| | | ```python |
| | |
| | | feats_dir="../DATA" #feature output dictionary |
| | | exp_dir="." |
| | | lang=zh |
| | | dumpdir=dump/fbank |
| | | feats_type=fbank |
| | | token_type=char |
| | | scp=wav.scp |
| | |
| | | done |
| | | wait |
| | | fi |
| | | |
| | | # Testing Stage |
| | | if [ ${stage} -le 4 ] && [ ${stop_stage} -ge 4 ]; then |
| | | echo "stage 4: Inference" |
| | | for dset in ${test_sets}; do |
| | | asr_exp=${exp_dir}/exp/${model_dir} |
| | | inference_tag="$(basename "${inference_config}" .yaml)" |
| | | _dir="${asr_exp}/${inference_tag}/${inference_asr_model}/${dset}" |
| | | _logdir="${_dir}/logdir" |
| | | if [ -d ${_dir} ]; then |
| | | echo "${_dir} is already exists. if you want to decode again, please delete this dir first." |
| | | exit 0 |
| | | fi |
| | | mkdir -p "${_logdir}" |
| | | _data="${feats_dir}/data/${dset}" |
| | | key_file=${_data}/${scp} |
| | | num_scp_file="$(<${key_file} wc -l)" |
| | | _nj=$([ $inference_nj -le $num_scp_file ] && echo "$inference_nj" || echo "$num_scp_file") |
| | | split_scps= |
| | | for n in $(seq "${_nj}"); do |
| | | split_scps+=" ${_logdir}/keys.${n}.scp" |
| | | done |
| | | # shellcheck disable=SC2086 |
| | | utils/split_scp.pl "${key_file}" ${split_scps} |
| | | _opts= |
| | | if [ -n "${inference_config}" ]; then |
| | | _opts+="--config ${inference_config} " |
| | | fi |
| | | ${infer_cmd} --gpu "${_ngpu}" --max-jobs-run "${_nj}" JOB=1:"${_nj}" "${_logdir}"/asr_inference.JOB.log \ |
| | | python -m funasr.bin.asr_inference_launch \ |
| | | --batch_size 1 \ |
| | | --ngpu "${_ngpu}" \ |
| | | --njob ${njob} \ |
| | | --gpuid_list ${gpuid_list} \ |
| | | --data_path_and_name_and_type "${_data}/${scp},speech,${type}" \ |
| | | --key_file "${_logdir}"/keys.JOB.scp \ |
| | | --asr_train_config "${asr_exp}"/config.yaml \ |
| | | --asr_model_file "${asr_exp}"/"${inference_asr_model}" \ |
| | | --output_dir "${_logdir}"/output.JOB \ |
| | | --mode asr \ |
| | | ${_opts} |
| | | |
| | | for f in token token_int score text; do |
| | | if [ -f "${_logdir}/output.1/1best_recog/${f}" ]; then |
| | | for i in $(seq "${_nj}"); do |
| | | cat "${_logdir}/output.${i}/1best_recog/${f}" |
| | | done | sort -k1 >"${_dir}/${f}" |
| | | fi |
| | | done |
| | | python utils/proce_text.py ${_dir}/text ${_dir}/text.proc |
| | | python utils/proce_text.py ${_data}/text ${_data}/text.proc |
| | | python utils/compute_wer.py ${_data}/text.proc ${_dir}/text.proc ${_dir}/text.cer |
| | | tail -n 3 ${_dir}/text.cer > ${_dir}/text.cer.txt |
| | | cat ${_dir}/text.cer.txt |
| | | done |
| | | fi |
| | |
| | | from funasr.modules.attention import MultiHeadedAttention |
| | | |
| | | |
| | | from funasr.train.abs_espnet_model import AbsESPnetModel |
| | | from funasr.models.base_model import FunASRModel |
| | | |
| | | |
| | | @torch.no_grad() |
| | | def calculate_all_attentions( |
| | | model: AbsESPnetModel, batch: Dict[str, torch.Tensor] |
| | | model: FunASRModel, batch: Dict[str, torch.Tensor] |
| | | ) -> Dict[str, List[torch.Tensor]]: |
| | | """Derive the outputs from the all attention layers |
| | | |
| | |
| | | from funasr.fileio.npy_scp import NpyScpWriter |
| | | from funasr.torch_utils.device_funcs import to_device |
| | | from funasr.torch_utils.forward_adaptor import ForwardAdaptor |
| | | from funasr.train.abs_espnet_model import AbsESPnetModel |
| | | from funasr.models.base_model import FunASRModel |
| | | |
| | | |
| | | @torch.no_grad() |
| | | def collect_stats( |
| | | model: AbsESPnetModel, |
| | | model: FunASRModel, |
| | | train_iter: DataLoader and Iterable[Tuple[List[str], Dict[str, torch.Tensor]]], |
| | | valid_iter: DataLoader and Iterable[Tuple[List[str], Dict[str, torch.Tensor]]], |
| | | output_dir: Path, |
| | |
| | | yield |
| | | |
| | | |
| | | class TimestampPredictor(AbsESPnetModel): |
| | | class TimestampPredictor(FunASRModel): |
| | | """ |
| | | Author: Speech Lab of DAMO Academy, Alibaba Group |
| | | """ |
| | |
| | | sond=DiarSondModel, |
| | | eend_ola=DiarEENDOLAModel, |
| | | ), |
| | | type_check=AbsESPnetModel, |
| | | type_check=FunASRModel, |
| | | default="sond", |
| | | ) |
| | | encoder_choices = ClassChoices( |