Update app.py

app.py

@@ -1,75 +1,107 @@
-# app.py — Your original working version + repetition_penalty=1.2 + ngram=3
+# app.py — Batch file transcription (up to 10 files)
 
 import os
-os.environ["OMP_NUM_THREADS"] = "1"
-os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "max_split_size_mb:128"
+import gc
+import zipfile
+import tempfile
 
 import gradio as gr
 import spaces
 from transformers import pipeline
 import torch
-import gc
+
+os.environ["OMP_NUM_THREADS"] = "1"
+os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "max_split_size_mb:128"
+
 
 # ——————————————————————————————
-# ZeroGPU worker – model loaded inside
+# ZeroGPU worker – model loaded once
 # ——————————————————————————————
 @spaces.GPU(duration=180)
-def transcribe_3min(audio_path):
-    if not audio_path:
-        return "Hlaðið upp hljóðskrá"
+def transcribe_files(audio_files):
+    if not audio_files:
+        return None, "Hlaðið upp hljóðskrám"
+
+    audio_files = audio_files[:10]
+
+    workdir = tempfile.mkdtemp()
+    outdir = os.path.join(workdir, "transcripts")
+    os.makedirs(outdir, exist_ok=True)
 
     pipe = pipeline(
-        "automatic-speech-recognition", 
-        #model="palli23/whisper-tiny-icelandic-distilled-v3",
-        #model = "palli23/whisper-tiny-distilled-spjallromur-polish-v3",
-        #odel = "palli23/whisper-tiny-distilled-spjallromur-polish-v5",
-        #model="palli23/whisper-tiny-distilled-samromur-spjallromur-polish",
-        #model="palli23/whisper-tiny-samromur-spjallromur",
+        "automatic-speech-recognition",
         model="palli23/whisper-small-sam_spjall",
         torch_dtype=torch.float16,
-        device=0,  # GPU inside @spaces.GPU
+        device=0,
     )
 
-    result = pipe(
-        audio_path,
-        chunk_length_s=30,
-        batch_size=8,
-        return_timestamps=False,  # ← no timestamps, as you want
-        generate_kwargs={
-            "num_beams": 5,  #var beam size 1
-            "repetition_penalty": 1.2,     # ← exactly what you asked for
-            "no_repeat_ngram_size": 3,     # ← exactly what you asked for
-            "temperature": 0.0,
-        }
-    )
+    for file in audio_files:
+        audio_path = file.name
+        base = os.path.splitext(os.path.basename(audio_path))[0]
+        txt_path = os.path.join(outdir, f"{base}.txt")
 
-    # Clean memory so ZeroGPU lives forever
+        result = pipe(
+            audio_path,
+            chunk_length_s=30,
+            batch_size=8,
+            return_timestamps=False,
+            generate_kwargs={
+                "num_beams": 5,
+                "repetition_penalty": 1.2,
+                "no_repeat_ngram_size": 3,
+                "temperature": 0.0,
+            },
+        )
+
+        with open(txt_path, "w", encoding="utf-8") as f:
+            f.write(result["text"].strip())
+
+    # Zip outputs
+    zip_path = os.path.join(workdir, "transcripts.zip")
+    with zipfile.ZipFile(zip_path, "w", zipfile.ZIP_DEFLATED) as z:
+        for fname in os.listdir(outdir):
+            z.write(os.path.join(outdir, fname), arcname=fname)
+
+    # Cleanup
     del pipe
     gc.collect()
     torch.cuda.empty_cache()
 
-    return result["text"]
+    return zip_path, "Lokið ✅"
+
 
 # ——————————————————————————————
-# UI – clean and simple
+# UI
 # ——————————————————————————————
 with gr.Blocks() as demo:
-    gr.Markdown("# Íslenskt ASR – 3 mínútur")
-    gr.Markdown("**palli23/whisper-small-sam_spjall** · mjög lágur WER · allt að 5 mín hljóð")
-    gr.Markdown("**Hafa samband:** pallinr1@protonmail.com")
+    gr.Markdown("# Íslenskt ASR – Batch (allt að 10 skrár)")
+    gr.Markdown(
+        "**palli23/whisper-small-sam_spjall** · sama stillingar · .wav / .mp3"
+    )
+
+    audio_in = gr.File(
+        label="Hlaðið upp allt að 10 .wav / .mp3 skrám",
+        file_types=[".wav", ".mp3"],
+        file_count="multiple",
+    )
 
-    audio_in = gr.Audio(type="filepath", label="Hlaðið upp .mp3 / .wav")
     btn = gr.Button("Transcribe", variant="primary", size="lg")
-    output = gr.Textbox(lines=25, label="Útskrift")
 
-    btn.click(fn=transcribe_3min, inputs=audio_in, outputs=output)
+    zip_out = gr.File(label="Niðurhal – transcripts.zip")
+    status = gr.Textbox(label="Staða", interactive=False)
+
+    btn.click(
+        fn=transcribe_files,
+        inputs=audio_in,
+        outputs=[zip_out, status],
+    )
+
 
 # ——————————————————————————————
-# Public launch
+# Launch
 # ——————————————————————————————
 demo.launch(
     share=True,
     server_name="0.0.0.0",
     server_port=7860,
-    auth=None
-)
+)