Fix: use fish-audio-sdk (not fishaudio), correct import

ingest.py

@@ -109,18 +109,18 @@ def transcribe_chunk_fish(chunk_path, api_key, start_offset_sec):
     Returns a list of segments with timestamps adjusted by start_offset_sec.
     Note: Fish Audio returns timestamps in milliseconds.
     """
-    from fishaudio import FishAudio
+    from fish_audio_sdk import Session, ASRRequest
 
     print(f"  → Fish Audio ASR: {chunk_path.name} (offset +{start_offset_sec}s)", flush=True)
 
-    client = FishAudio(api_key=api_key)
-    with open(chunk_path, "rb") as f:
-        result = client.asr.transcribe(audio=f.read(), language="en")
+    with Session(api_key) as session:
+        with open(chunk_path, "rb") as f:
+            result = session.asr(ASRRequest(audio=f.read(), language="en"))
 
     # Fish Audio returns timestamps in milliseconds — convert to seconds
     # and adjust by the chunk's start offset in the full audio
     segments = []
-    for seg in result.segments:
+    for seg in (result.segments or []):
         segments.append({
             "start": seg.start / 1000.0 + start_offset_sec,
             "end": seg.end / 1000.0 + start_offset_sec,
@@ -128,7 +128,7 @@ def transcribe_chunk_fish(chunk_path, api_key, start_offset_sec):
         })
 
     # Fallback: if no segments, use full text as one block
-    if not segments and result.text:
+    if not segments and getattr(result, "text", None):
         segments.append({
             "start": start_offset_sec,
             "end": start_offset_sec + 60,