funasr1.0

游雁

2024-01-15 a035d68e860ea6decdf422c0fc04eda4fc4de397

 funasr/bin/inference.py

@@ -115,7 +115,7 @@
        vad_model = kwargs.get("vad_model", None)
        vad_kwargs = kwargs.get("vad_model_revision", None)
        if vad_model is not None:
            print("build vad model")
            logging.info("Building VAD model.")
            vad_kwargs = {"model": vad_model, "model_revision": vad_kwargs}
            vad_model, vad_kwargs = self.build_model(**vad_kwargs)

@@ -123,6 +123,7 @@
        punc_model = kwargs.get("punc_model", None)
        punc_kwargs = kwargs.get("punc_model_revision", None)
        if punc_model is not None:
            logging.info("Building punc model.")
            punc_kwargs = {"model": punc_model, "model_revision": punc_kwargs}
            punc_model, punc_kwargs = self.build_model(**punc_kwargs)

@@ -130,6 +131,7 @@
        spk_model = kwargs.get("spk_model", None)
        spk_kwargs = kwargs.get("spk_model_revision", None)
        if spk_model is not None:
            logging.info("Building SPK model.")
            spk_kwargs = {"model": spk_model, "model_revision": spk_kwargs}
            spk_model, spk_kwargs = self.build_model(**spk_kwargs)
            self.cb_model = ClusterBackend()
@@ -173,7 +175,7 @@
        # build tokenizer
        tokenizer = kwargs.get("tokenizer", None)
        if tokenizer is not None:
            tokenizer_class = tables.tokenizer_classes.get(tokenizer.lower())
            tokenizer_class = tables.tokenizer_classes.get(tokenizer)
            tokenizer = tokenizer_class(**kwargs["tokenizer_conf"])
            kwargs["tokenizer"] = tokenizer
            kwargs["token_list"] = tokenizer.token_list
@@ -184,13 +186,13 @@
        # build frontend
        frontend = kwargs.get("frontend", None)
        if frontend is not None:
            frontend_class = tables.frontend_classes.get(frontend.lower())
            frontend_class = tables.frontend_classes.get(frontend)
            frontend = frontend_class(**kwargs["frontend_conf"])
            kwargs["frontend"] = frontend
            kwargs["input_size"] = frontend.output_size()
        
        # build model
        model_class = tables.model_classes.get(kwargs["model"].lower())
        model_class = tables.model_classes.get(kwargs["model"])
        model = model_class(**kwargs, **kwargs["model_conf"], vocab_size=vocab_size)
        model.eval()
        model.to(device)
@@ -403,7 +405,7 @@
                spk_embedding = result['spk_embedding']
                labels = self.cb_model(spk_embedding, oracle_num=self.preset_spk_num)
                del result['spk_embedding']
                sv_output = postprocess(all_segments, None, labels, spk_embedding)
                sv_output = postprocess(all_segments, None, labels, spk_embedding.cpu())
                if self.spk_mode == 'vad_segment':
                    sentence_list = []
                    for res, vadsegment in zip(restored_data, vadsegments):
@@ -441,7 +443,7 @@
        # build frontend
        frontend = kwargs.get("frontend", None)
        if frontend is not None:
            frontend_class = tables.frontend_classes.get(frontend.lower())
            frontend_class = tables.frontend_classes.get(frontend)
            frontend = frontend_class(**kwargs["frontend_conf"])

        self.frontend = frontend