update

游雁

2024-07-17 a836eca98e30fa67d45167dac40f359ae42d42ec

 funasr/auto/auto_model.py

@@ -315,7 +315,7 @@
            speed_stats["rtf"] = f"{(time_escape) / batch_data_time:0.3f}"
            description = f"{speed_stats}, "
            if pbar:
                pbar.update(1)
                pbar.update(end_idx - beg_idx)
                pbar.set_description(description)
            time_speech_total += batch_data_time
            time_escape_total += time_escape
@@ -339,7 +339,9 @@
        #  FIX(gcf): concat the vad clips for sense vocie model for better aed
        if kwargs.get("merge_vad", False):
            for i in range(len(res)):
                res[i]["value"] = merge_vad(res[i]["value"], kwargs.get("merge_length", 15000))
                res[i]["value"] = merge_vad(
                    res[i]["value"], kwargs.get("merge_length_s", 15) * 1000
                )

        # step.2 compute asr model
        model = self.model
@@ -380,6 +382,9 @@
            if len(sorted_data) > 0 and len(sorted_data[0]) > 0:
                batch_size = max(batch_size, sorted_data[0][0][1] - sorted_data[0][0][0])

            if kwargs["device"] == "cpu":
                batch_size = 0

            beg_idx = 0
            beg_asr_total = time.time()
            time_speech_total_per_sample = speech_lengths / 16000