Update app.py

app.py

@@ -18,7 +18,7 @@ except ImportError:
 import yt_dlp  # Added import for yt-dlp
 
 MODEL_NAME = "NbAiLab/nb-whisper-large"
-lang = "no"
+#lang = "no"
 
 max_audio_length= 30 * 60
 
@@ -28,7 +28,7 @@ device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 print(f"Bruker enhet: {device}")
 
 @spaces.GPU(duration=60 * 2)
-def pipe(file, return_timestamps=False,lang="no"):
+def pipe(file, return_timestamps=False,lang_nn=False):
     asr = pipeline(
         task="automatic-speech-recognition",
         model=MODEL_NAME,
@@ -38,12 +38,19 @@ def pipe(file, return_timestamps=False,lang="no"):
         torch_dtype=torch.float16,
         model_kwargs={"attn_implementation": "flash_attention_2", "num_beams": 5} if FLASH_ATTENTION else {"attn_implementation": "sdpa", "num_beams": 5},
     )
-    asr.model.config.forced_decoder_ids = asr.tokenizer.get_decoder_prompt_ids(
-        language=lang,
-        task="transcribe",
-        no_timestamps=not return_timestamps,
-    )
-    return asr(file, return_timestamps=return_timestamps, batch_size=24)
+    if not lang_nn:
+        asr.model.config.forced_decoder_ids = asr.tokenizer.get_decoder_prompt_ids(
+            language="no",
+            task="transcribe",
+            no_timestamps=not return_timestamps,
+        )
+    else:
+        asr.model.config.forced_decoder_ids = asr.tokenizer.get_decoder_prompt_ids(
+            language="nn",
+            task="transcribe",
+            no_timestamps=not return_timestamps,
+        )
+    return asr(file, return_timestamps=return_timestamps, lang_nn=lang_nn, batch_size=24)
 
 def format_output(text):
     # Add a line break after ".", "!", ":", or "?" unless part of sequences like "..."
@@ -84,10 +91,10 @@ def transcribe(file, return_timestamps=False,lang_nn=False):
             formatted_text = "<br>".join(text)
     else:
         if not return_timestamps:
-            text = pipe(file_to_transcribe,lang="nn")["text"]
+            text = pipe(file_to_transcribe,lang_nn=True)["text"]
             formatted_text = format_output(text)
         else:
-            chunks = pipe(file_to_transcribe, return_timestamps=True,lang="nn")["chunks"]
+            chunks = pipe(file_to_transcribe, return_timestamps=True,lang_nn=True)["chunks"]
             text = []
             for chunk in chunks:
                 start_time = time.strftime('%H:%M:%S', time.gmtime(chunk["timestamp"][0])) if chunk["timestamp"][0] is not None else "??:??:??"