From 51882b5065976945e3a86b03993262d6433fbbb6 Mon Sep 17 00:00:00 2001
From: jamiepine <32987599+jamiepine@users.noreply.github.com>
Date: Sat, 3 Oct 2026 09:39:10 +0000
Subject: [PATCH] fix(backend): keep the pad-to-30s Whisper path for clips
 under 30s

With padding="longest" the encoder rejects inputs shorter than 3000 mel
frames when generate() has to detect the language first, so every short
clip without a forced language failed with "Whisper expects the mel input
features to be of length 3000". Use the long-form feature-extractor and
generate() options only when the audio exceeds one 30s window.
---
 backend/backends/pytorch_backend.py | 22 +++++++++++++++++-----
 1 file changed, 17 insertions(+), 5 deletions(-)

diff --git a/backend/backends/pytorch_backend.py b/backend/backends/pytorch_backend.py
index 012fd85e..af0f4a78 100644
--- a/backend/backends/pytorch_backend.py
+++ b/backend/backends/pytorch_backend.py
@@ -348,13 +348,24 @@ class PyTorchSTTBackend:
             # are required for long-form transcription. Without them the
             # feature extractor silently truncates to 30s (Whisper's native
             # window) and audio past that point is dropped.
+            #
+            # Only use them for audio longer than one 30s window. For shorter
+            # clips the default pad-to-30s path is needed: with
+            # padding="longest" the encoder rejects the shorter mel input
+            # ("expects the mel input features to be of length 3000") when
+            # generate() runs language detection, i.e. whenever no language
+            # is forced.
+            is_long_form = len(audio) > 30 * 16000
+            processor_kwargs = (
+                {"truncation": False, "padding": "longest", "return_attention_mask": True}
+                if is_long_form
+                else {}
+            )
             inputs = self.processor(
                 audio,
                 sampling_rate=16000,
                 return_tensors="pt",
-                truncation=False,
-                padding="longest",
-                return_attention_mask=True,
+                **processor_kwargs,
             )
             inputs = inputs.to(self.device)
 
@@ -369,12 +380,13 @@ class PyTorchSTTBackend:
             if language:
                 generate_kwargs["language"] = language
                 generate_kwargs["task"] = "transcribe"
+            if is_long_form:
+                generate_kwargs["attention_mask"] = inputs["attention_mask"]
+                generate_kwargs["return_timestamps"] = True
 
             with torch.no_grad():
                 predicted_ids = self.model.generate(
                     inputs["input_features"],
-                    attention_mask=inputs["attention_mask"],
-                    return_timestamps=True,
                     **generate_kwargs,
                 )