[Export] fix bug

wanchen.swc

2023-03-30 999c0837b3268c51dc94cec14cd837b4c2ef2d63

[Export] fix bug

 funasr/runtime/python/libtorch/funasr_torch/paraformer_bin.py

@@ -46,6 +46,7 @@
        )
        self.ort_infer = torch.jit.load(model_file)
        self.batch_size = batch_size
        self.device_id = device_id
        self.plot_timestamp_to = plot_timestamp_to
        self.pred_bias = pred_bias

@@ -58,11 +59,13 @@
            end_idx = min(waveform_nums, beg_idx + self.batch_size)
            feats, feats_len = self.extract_feat(waveform_list[beg_idx:end_idx])
            try:
                if int(device_id) != -1:
                    feats = feats.cuda()
                    feats_len = feats_len.cuda()
                outputs = self.ort_infer(feats, feats_len)
                am_scores, valid_token_lens = outputs[0], outputs[1]
                with torch.no_grad():
                    if int(self.device_id) == -1:
                        outputs = self.ort_infer(feats, feats_len)
                        am_scores, valid_token_lens = outputs[0], outputs[1]
                    else:
                        outputs = self.ort_infer(feats.cuda(), feats_len.cuda())
                        am_scores, valid_token_lens = outputs[0].cpu(), outputs[1].cpu()
                if len(outputs) == 4:
                    # for BiCifParaformer Inference
                    us_alphas, us_peaks = outputs[2], outputs[3]