Skip to content

Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

Gcf (#1605)

gaochangfeng851e3e3ef83d
diff --git a/funasr/models/sense_voice/whisper_lib/decoding.py b/funasr/models/sense_voice/whisper_lib/decoding.py
index b3fce7ef2..2239b642a 100644
--- a/funasr/models/sense_voice/whisper_lib/decoding.py
+++ b/funasr/models/sense_voice/whisper_lib/decoding.py
@@ -10,6 +10,8 @@ from torch.distributions import Categorical
from .audio import CHUNK_LENGTH
from .tokenizer import Tokenizer, get_tokenizer
from .utils import compression_ratio
+from funasr.models.transformer.utils.nets_utils import to_device
+
if TYPE_CHECKING:
from .model import Whisper
@@ -58,6 +60,10 @@ def detect_language(
# x = torch.tensor([[tokenizer.sot]] * n_audio).to(mel.device) # [n_audio, 1]
if x is None:
x = torch.tensor([tokenizer.encode(initial_prompt, allowed_special="all")] * n_audio).to(mel.device) # [n_audio, 1]
+
+ else:
+ x = x.to(mel.device)
+
logits = model.logits(x[:,:-1], mel)[:, -1]
# collect detected languages; suppress all non-language tokens
mask = torch.ones(logits.shape[-1], dtype=torch.bool)