Transcripts

Originally, the intention was to automate ingest of recorded files to Dragon NaturallySpeaking, and to produce actual transcripts of recorded audio notes. However, the speech-to-text engine doesn't do a good job, perhaps because of the car noise or just my voice. Instead, the new goal is to produce "audio transcripts," with car noise reduced and silence removed.

Components

• LG Tone Infim HBS920 Bluetooth headset
• Sony Xperia XA2 or other Android handset
• axet android-audio-recorder
• Audacity
• 15-sec WAV file (48kHz) of car noise; not saved in Git

Workflow

1. Pair & connect headset to phone, to use while driving
2. Configure Audio Recorder app with the following settings:
   • Storage Path: default, under "Android"
   • Recording Source: Bluetooth
   • Sample Rate: 48kHz
   • Encoding: .flac
   • Mode: Mono (default)
   • Name Format: 2020-01-20 13.58.41.flac
   • Bandpass Voice Filter: enabled
   • Recording Volume: 100%
   • Skip Silence: disabled
   • Encoding on Fly: disabled
   • Pause During Calls: enabled
   • Silence Mode: disabled
   • Lockscreen Controls: disabled
3. Start recording; don't worry about leaving silence to think
4. Stop recording
5. Later, once off the road, rename & upload the file (e.g., w/ QFile)
6. Download the file to a computer with Audacity, to /tmp/macro-input
7. Copy CarNoise.wav (not included) into /tmp/macro-input as well, and rename it so that it will be sorted first (e.g., 000-CarNoise.wav)
8. Set up the Remove Car Noise macro to match the included PNG
9. Run the macro on Files..., and select all files in /tmp/macro-input -- note that Audacity seems to have problems when more than 100 files are selected
10. Run the adjust_mtimes.sh script with no arguments
11. Upload the xx_processed.ogg files from /tmp/macro-output to NAS