diff --git a/whisper/transcribe.py b/whisper/transcribe.py
index 0a4cc36..ee55060 100644
--- a/whisper/transcribe.py
+++ b/whisper/transcribe.py
@@ -52,6 +52,7 @@ def transcribe(
     append_punctuations: str = "\"'.。,，!！?？:：”)]}、",
     clip_timestamps: Union[str, List[float]] = "0",
     hallucination_silence_threshold: Optional[float] = None,
+    mel_spectrogram_device: Optional[Union[str, torch.device]] = None,
     **decode_options,
 ):
     """
@@ -119,6 +120,9 @@ def transcribe(
         When word_timestamps is True, skip silent periods longer than this threshold (in seconds)
         when a possible hallucination is detected
 
+    mel_spectrogram_device: Optional[Union[str, torch.device]]
+        If given, the audio tensor is moved to this device before STFT
+
     Returns
     -------
     A dictionary containing the resulting text ("text") and segment-level details ("segments"), and
@@ -136,7 +140,9 @@ def transcribe(
         decode_options["fp16"] = False
 
     # Pad 30-seconds of silence to the input audio, for slicing
-    mel = log_mel_spectrogram(audio, model.dims.n_mels, padding=N_SAMPLES)
+    mel = log_mel_spectrogram(
+        audio, model.dims.n_mels, padding=N_SAMPLES, device=mel_spectrogram_device
+    )
     content_frames = mel.shape[-1] - N_FRAMES
     content_duration = float(content_frames * HOP_LENGTH / SAMPLE_RATE)