Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.
Python asraudiochineseemotion-recognitionfunasrmcp-servermultilingual-asropenai-compatible-apiparaformerpunctuationpytorchreal-time-asrspeaker-diarizationspeech-recognitionspeech-to-textstreaming-asrtranscriptionvllmvoice-activity-detectionwhisper-alternative
fix bug
游雁6e0d1388264f
diff --git a/funasr/datasets/openai_datasets/datasets.py b/funasr/datasets/openai_datasets/datasets.pyindex 9c74ef188..e13c15618 100644--- a/funasr/datasets/openai_datasets/datasets.py+++ b/funasr/datasets/openai_datasets/datasets.py@@ -211,14 +211,15 @@ class OpenAIDataset(torch.utils.data.Dataset):if self.batch_type != "example":b, t = outputs["input_ids"].shape- if b * t > self.batch_size * self.batch_size_scale_ratio_max:- beg = torch.randint(0, 2, ()).item()- if b < 2:- beg = 0+ if b > 1 and b * t > self.batch_size * self.batch_size_scale_ratio_max:+ # beg = torch.randint(0, 2, ()).item()+ # if b < 2:+ # beg = 0logging.info(- f"Warning, b * t: {b * t} > {self.batch_size_scale_ratio_max} * {self.batch_size}, b: {b}, t: {t}, drop half data {idx}th, beg:{beg}"+ f"Warning, b*t: {b}*{t}={b * t} > batch_size*relax: {self.batch_size_scale_ratio_max}*{self.batch_size}={self.batch_size_scale_ratio_max*self.batch_size}, drop half data {idx}th, beg:{beg}")- samples = samples[beg : beg + b : 2]+ # samples = samples[beg : beg + b : 2]+ samples = samples[:-1]continuebreak