feat: Handle markdown in model output and fix dependencies

main.py

@@ -173,6 +173,16 @@ def split_audio_webrtcvad(audio_segment, min_silence_len):
     return final_chunks
 
 
+def _trim_markdown_fences(text: str) -> str:
+    """Removes markdown code block fences from a string."""
+    text = text.strip()
+    if text.startswith("```") and text.endswith("```"):
+        text = text[3:-3].strip()
+        if text.startswith("json"):
+            text = text[4:].strip()
+    return text
+
+
 def validate_and_correct_segments(segments_from_api, chunk_duration_ms):
     corrected_segments, last_corrected_end_ms = [], 0
     for seg in segments_from_api:
@@ -234,7 +244,8 @@ def transcribe_chunk(chunk_audio, api_key, system_prompt, pydantic_schema, model
             response_schema=list[pydantic_schema]
         )
     )
-    return json.loads(response.text), None
+    cleaned_text = _trim_markdown_fences(response.text)
+    return json.loads(cleaned_text), None
 
 
 def transcribe_chunk_plain_text(chunk_audio, api_key, system_prompt, model_name, user_prompt):
@@ -260,7 +271,8 @@ def transcribe_chunk_plain_text(chunk_audio, api_key, system_prompt, model_name,
             contents=contents,
             config=types.GenerateContentConfig(system_instruction=system_prompt)
         )
-        return response.text, None
+        cleaned_text = _trim_markdown_fences(response.text)
+        return cleaned_text, None
     except Exception as e:
         logging.error(f"Error in transcribe_chunk_plain_text: {e}")
         return None, str(e)

requirements.txt

@@ -1,9 +1,10 @@
-fastapi
-uvicorn[standard]
-python-multipart
-google-genai
-pydub
-PyYAML
-Jinja2
-pydantic
-webrtcvad
+fastapi
+uvicorn[standard]
+python-multipart
+google-genai
+pydub
+PyYAML
+Jinja2
+pydantic
+webrtcvad
+setuptools