"Thank you" is being added to transcriptions generated after recording.

[jjordansd]

How do I disable the "Thank you" from being added to transcriptions generated after recording? When using the Command+Shift+; to start recording and end recording the transcription is generated from the recording and a "Thank you" is appended to the transcribed text. I see transcribe-rs is used and hoping someone can help here.

I'd like to disable this but I'm not sure how the transcription is generated or processed. Here are some of my settings:

custom Preview shows:
ffmpeg -i input.wav -af silenceremove=start_periods=1:start_duration=0.1:start_threshold=-50dB:detection=peak,aformat=s16:16000:1 -c:a libopus -b:a 32k -ar 16000 -ac 1 -compression_level 10 output.opus

[Leftium]

"Thank you" (or similar hallucinations) are often added to transcription output when there is silence in the audio input.

• The training data likely included videos with subtitles.
• Some people who create subtitles ("fansubs") insert messages during periods of silence, especially at the end and/or beginning.
• So the AI (transcription model like Whisper) associates silence with these inserted messages.

---

Removing silence (more aggressively) may help with unwanted "thank you"s. Here are custom options optimized for transcription:

ffmpeg -i input.wav -af "silenceremove=start_periods=1:start_duration=0.1:start_threshold=-45dB:stop_periods=-1:stop_duration=0. 25:stop_threshold=-35dB:detection=rms,volume=1.2,atempo=2.0,aformat=s16:16000:1" -c:a libopus -b:a 32k -ar 16000 -ac 1 -compression_level 10 output.opus

Note: there are some bug reports that compression is not happening before transcription. That would mean this silence removal would also not be happening. So changing the options may not have any effect.

---

Explanation of each setting above:

# Silence removal at start
silenceremove=
  start_periods=1:              # Remove 1 silence segment at beginning
  start_duration=0.1:           # Remove silence longer than 0.1 seconds at start
  start_threshold=-45dB:        # Audio below -45dB considered silence at start

# Silence removal throughout audio  
  stop_periods=-1:              # Remove ALL silence segments throughout (-1 = unlimited)
  stop_duration=0.25:           # Remove silence gaps longer than 0.25 seconds
  stop_threshold=-35dB:         # Audio below -35dB considered silence (less sensitive than start)
  detection=rms,                # Use RMS detection (smooth, good for speech)

# Audio enhancement
volume=1.2,                     # Boost volume by 20% for better clarity

# Speed adjustment
atempo=2.0,                     # Speed up audio by 2x (preserves pitch, no chipmunk effect)

# Format specification
aformat=s16:16000:1             # Force 16-bit signed, 16kHz sample rate, mono