Buzz Transcription - Revision history

Justinaquino: Point Taglish audio + report to wiki.gi7b.org copies (mp3 not allowed here; 2 MB upload cap)

2026-08-09T06:20:05Z

Point Taglish audio + report to wiki.gi7b.org copies (mp3 not allowed here; 2 MB upload cap)

← Older revision		Revision as of 06:20, 9 August 2026
Line 222:		Line 222:
	A Taglish (Tagalog + English code-switching) test recording, transcribed automatically by Buzz's Folder Watch during testing (now disabled — see the [[#Folder Watch (Automatic Transcription)\|Folder Watch]] warning above):		A Taglish (Tagalog + English code-switching) test recording, transcribed automatically by Buzz's Folder Watch during testing (now disabled — see the [[#Folder Watch (Automatic Transcription)\|Folder Watch]] warning above):

	* '''Audio:''' [[:~~File~~:~~Taglish Test 260809~~.mp3\|Taglish Test 260809.mp3]]		* '''Audio:''' [https://wiki.gi7b.org/index.php/Special:FilePath/Taglish_Test_260809.mp3 Taglish Test 260809.mp3] — hosted on wiki.gi7b.org (this wiki does not allow audio uploads)
	* '''Evaluation report:''' [[:~~File~~:~~Taglish Transcription and Evaluation Report 260809~~.docx\|Taglish Transcription and Evaluation Report 260809.docx]]		* '''Evaluation report:''' [https://wiki.gi7b.org/index.php/Special:FilePath/Taglish_Transcription_and_Evaluation_Report_260809.docx Taglish Transcription and Evaluation Report 260809.docx] — hosted on wiki.gi7b.org (this wiki caps uploads at 2 MB)

	The report grades the result '''95/100 (Excellent)''' at ~96 % word accuracy. Code-switching between Filipino and English (EDSA, client, project, presentation, brain cells, lunch, work) is handled seamlessly. Noted deviations are typical ASR behavior: punctuation is collapsed into sentence breaks, "Almost two hours" became "Almost 2 hours", "na-stuck" → "nakastuck", "kailangan nating" → "kailan nating".		The report grades the result '''95/100 (Excellent)''' at ~96 % word accuracy. Code-switching between Filipino and English (EDSA, client, project, presentation, brain cells, lunch, work) is handled seamlessly. Noted deviations are typical ASR behavior: punctuation is collapsed into sentence breaks, "Almost two hours" became "Almost 2 hours", "na-stuck" → "nakastuck", "kailangan nating" → "kailan nating".

Justinaquino: Mirror from wiki.gi7b.org rev 4963 (2026-08-09): Folder Watch warning, Taglish example, screenshots

2026-08-09T06:10:18Z

Mirror from wiki.gi7b.org rev 4963 (2026-08-09): Folder Watch warning, Taglish example, screenshots

← Older revision		Revision as of 06:10, 9 August 2026
Line 156:		Line 156:

	''Investigation date: 2026-04-25. System: Ubuntu 24.04, AMD RX 7600, Buzz 1.4.4 (Flatpak).''		''Investigation date: 2026-04-25. System: Ubuntu 24.04, AMD RX 7600, Buzz 1.4.4 (Flatpak).''

			== Folder Watch (Automatic Transcription) ==

			<div style="border:1px solid #a00;background:#fee;padding:0.5em 1em;">'''Do not use Folder Watch.''' It is disruptive at default settings: every audio file dropped into the watched folder is transcribed automatically without asking, silently occupying the GPU/CPU and flooding the folder with output files — and with ''Delete processed files'' ticked, the source audio is deleted afterwards. It is disabled by default and should stay that way. Use [[#File Transcription (Meeting Recording)\|File → Import Media]] instead. The settings below are documented for reference only (as configured during testing on 2026-08-09).</div>

			[[File:Buzz Folder Watch tab.png\|thumb\|400px\|right\|Folder Watch tab: whisper.cpp backend, Large-V3, advanced options unchecked]]
			[[File:Buzz Folder Watch tab full.png\|thumb\|400px\|right\|Folder Watch tab — Enable folder watch with input/output folders set to /home/justin/Music]]

			=== Settings ===

			In '''Preferences → Folder Watch''':

			{\| class="wikitable"
			\|-
			! Setting !! Value !! Notes
			\|-
			\| Enable folder watch \|\| '''leave unticked''' \|\| ticking it starts auto-transcribing everything dropped into the input folder
			\|-
			\| Input folder \|\| <code>/home/justin/Music</code> \|\| drop recordings here
			\|-
			\| Output folder \|\| <code>/home/justin/Music</code> \|\| same folder — files stay put
			\|-
			\| Delete processed files \|\| '''unchecked''' \|\| '''ticking this deletes the source audio after transcription''' — keep it unticked
			\|-
			\| Model \|\| whisper.cpp — Large-V3 \|\| engine choice matters, see below
			\|-
			\| Task \|\| Transcribe \|\|
			\|-
			\| Language \|\| Detect Language \|\|
			\|-
			\| Export \|\| TXT \|\| SRT / VTT also available
			\|-
			\| Word-level timings \|\| '''unchecked''' \|\|
			\|-
			\| Extract speech \|\| '''unchecked''' \|\| '''do not check — crashes on long files''' (see [[#OOM Crash with Large Files\|OOM Crash with Large Files]])
			\|}

			=== Use whisper.cpp instead of the default engine ===

			[[File:Buzz model dropdown.png\|thumb\|400px\|right\|Model dropdown: whisper.cpp and Faster Whisper]]

			The model picker lists two engines: '''whisper.cpp''' and '''Faster Whisper'''. Faster Whisper is the default — switch to '''whisper.cpp''':

			* whisper.cpp uses the <code>ggml-*.bin</code> models already cached under <code>~/.var/app/io.github.chidiwilliams.Buzz/cache/Buzz/models/</code> (Flatpak path; <code>~/snap/buzz/…</code> on snap installs) — no extra download
			* it runs through the bundled Vulkan <code>whisper-cli</code>, so GPU acceleration applies (see [[#GPU Acceleration\|GPU Acceleration]])

			[[File:Buzz Models tab whisper.cpp.png\|thumb\|400px\|right\|Models tab, whisper.cpp group: Large-V3, Large-V3-Turbo, Medium.en, Large and Large-v2 downloaded]]

			In the Models tab, the whisper.cpp group shows the downloaded sizes — Large-V3 and Large-V3-Turbo are the useful ones for meetings (Large-V3-Turbo: best quality/speed trade-off, see [[#How to Transcribe a Meeting\|Recommended Settings]]).

			=== Leave the advanced options unchecked ===

			[[File:Buzz transcription dialog unchecked.png\|thumb\|400px\|right\|Transcription dialog: neither Word-level timings nor Extract speech checked]]

			In the transcription dialog (shown when folder watch picks up a file) both '''Word-level timings''' and '''Extract speech''' are left unchecked. Checking '''Extract speech''' — the obvious option for a meeting recording — triggers the Demucs pre-processing step that OOM-crashes the app on long files (22 GB RSS kill, [[#OOM Crash with Large Files\|see above]]). Keep both unchecked.

			=== Result ===

			Transcriptions are written next to the source as <code><original name> (transcribed on <date>).txt</code>, e.g. <code>2026-08-08 20-56-20 3rd Apocalypse Bob's ADND1E (transcribed on 09-Aug-2026 01-50-32).txt</code>.

			Config verified 2026-08-09 in <code>~/snap/buzz/570/.config/Buzz.conf</code> (snap install): <code>folder_watch\enabled</code> = <code>false</code> (off — recommended), <code>folder_watch\input_folder</code> = <code>/home/justin/Music</code>, <code>folder_watch\output_directory</code> = <code>/home/justin/Music</code>, model Whisper.cpp / large-v3, and <code>delete_processed_files</code>, <code>extract_speech</code>, <code>word_level_timings</code> all <code>false</code>.

			== Example: Taglish Test (2026-08-09) ==

			A Taglish (Tagalog + English code-switching) test recording, transcribed automatically by Buzz's Folder Watch during testing (now disabled — see the [[#Folder Watch (Automatic Transcription)\|Folder Watch]] warning above):

			* '''Audio:''' [[:File:Taglish Test 260809.mp3\|Taglish Test 260809.mp3]]
			* '''Evaluation report:''' [[:File:Taglish Transcription and Evaluation Report 260809.docx\|Taglish Transcription and Evaluation Report 260809.docx]]

			The report grades the result '''95/100 (Excellent)''' at ~96 % word accuracy. Code-switching between Filipino and English (EDSA, client, project, presentation, brain cells, lunch, work) is handled seamlessly. Noted deviations are typical ASR behavior: punctuation is collapsed into sentence breaks, "Almost two hours" became "Almost 2 hours", "na-stuck" → "nakastuck", "kailangan nating" → "kailan nating".

Justinaquino: Add Buzz meeting transcription guide + Vulkan GPU & OOM investigation (2026-04-25)

2026-04-24T18:43:50Z

Add Buzz meeting transcription guide + Vulkan GPU & OOM investigation (2026-04-25)

New page

Buzz is an open-source, offline-capable audio/meeting transcription tool powered by OpenAI Whisper (via whisper.cpp). On this system it is installed as a Flatpak and runs with Vulkan GPU acceleration on the AMD RX 7600.

== How to Transcribe a Meeting ==

=== Prerequisites ===
* Buzz installed (Flatpak: <code>io.github.chidiwilliams.Buzz</code>)
* A model downloaded — recommended: <code>ggml-large-v3-turbo</code> for quality, <code>ggml-tiny</code> for speed
* Models are stored at <code>~/.var/app/io.github.chidiwilliams.Buzz/cache/Buzz/models/</code>

=== File Transcription (Meeting Recording) ===
# Launch Buzz from the application menu or run: <code>flatpak run io.github.chidiwilliams.Buzz</code>
# Go to '''File → Import Media''' and select your meeting recording (MP3, WAV, M4A, etc.)
# In the transcription dialog:
#* '''Model''': Select <code>large-v3-turbo</code> (best quality for meetings)
#* '''Task''': Transcribe
#* '''Language''': Select your language or leave on Auto
#* '''Extract Speech''': '''Leave unchecked''' for files longer than ~30 minutes (see [[#OOM Crash with Large Files|OOM Crash]] below)
#* '''Output''': SRT, VTT, or TXT depending on your need
# Click '''Transcribe'''
# When complete, the transcript appears in the main window and is saved alongside the source file

=== Live Transcription (Real-Time) ===
# Go to '''File → Record and Transcribe'''
# Select your microphone input
# Choose model and language
# Click '''Record''' — transcription appears live on screen
# Stop recording when done; export via '''File → Export'''

=== Recommended Settings for Meetings ===
{| class="wikitable"
|-
! Setting !! Value !! Reason
|-
| Model || large-v3-turbo || Best accuracy; 1.6 GB VRAM, runs on RX 7600
|-
| Extract Speech || '''Off''' for files >30 min || Demucs uses 8–15 GB RAM for long files (OOM risk)
|-
| Language || Set explicitly || Faster than auto-detect
|-
| Task || Transcribe || Use Translate only if you need English output from another language
|}

=== Before Each Session: Clear Swap ===
Prior transcription sessions can leave stale pages in swap. Run this before starting a long transcription:
<pre>sudo swapoff -a && sudo swapon -a</pre>
This is safe when free RAM exceeds swap used (typical on this system: 15+ GB free, 8 GB swap).

== GPU Acceleration ==

Buzz's Flatpak installation includes a Vulkan-enabled <code>whisper-cli</code>. On this system, GPU inference is automatically active — no configuration needed.

* GPU in use: '''AMD RX 7600 (RDNA3, RADV NAVI33)'''
* Confirmed by: <code>whisper_backend_init_gpu: using Vulkan0 backend</code> in runtime output
* Buzz checks for Vulkan at startup (<code>IS_VULKAN_SUPPORTED</code>) and omits the <code>--no-gpu</code> flag when found

To force CPU inference (for debugging):
<pre>flatpak override --user io.github.chidiwilliams.Buzz --env=BUZZ_FORCE_CPU=true</pre>

To revert:
<pre>flatpak override --user --reset io.github.chidiwilliams.Buzz</pre>

== OOM Crash with Large Files ==

=== Symptom ===
When selecting <code>ggml-large-v3-turbo</code> in Buzz and starting a transcription on a long file, the application crashes. VRAM usage never visibly climbs before the crash.

=== Root Cause ===
The crash is caused by the '''Extract Speech (Demucs)''' pre-processing step, not the large model.

Demucs is a PyTorch-based music source separation model that strips background noise from audio before passing it to the transcription engine. It processes raw PCM audio at full float32 precision entirely in RAM:

{| class="wikitable"
|-
! Audio Duration !! Compressed Size !! RAM Required (Demucs)
|-
| 1 hour MP3 || ~60 MB || 500 MB – 1 GB
|-
| 3–4 hour session || ~220 MB || 8–15 GB peak
|}

The combination that caused the crash:
# Swap already exhausted from previous sessions
# Extract Speech (Demucs) triggered on a ~220 MB MP3 (~3–4 hours of audio)
# Python RAM usage exceeded available RAM + swap ceiling
# OOM killer terminated the process at 22 GB RSS
# <code>whisper-cli</code> was never launched → no VRAM activity observed

=== Why VRAM Never Climbed ===
<code>whisper-cli</code> runs as a subprocess of the Python GUI. The OOM kill happened during Demucs pre-processing — <code>whisper-cli</code> was never started, so no VRAM allocation occurred.

From the Buzz log at crash time:
<pre>~/.var/app/io.github.chidiwilliams.Buzz/.local/state/Buzz/log/logs.txt</pre>
The last entry was <code>Will extract speech</code> — the log never reached <code>Starting whisper file transcription</code>.

=== Fix ===
'''Uncheck "Extract Speech"''' in the Buzz transcription dialog for any file longer than ~30 minutes. The <code>large-v3-turbo</code> model has built-in noise tolerance that makes Demucs pre-processing unnecessary in most cases.

Optionally, grow the swapfile if you need Extract Speech for shorter files:
<pre>sudo swapoff /swap.img
sudo fallocate -l 16G /swap.img
sudo mkswap /swap.img
sudo swapon /swap.img</pre>

== Environment Variables ==

{| class="wikitable"
|-
! Variable !! Default !! Effect
|-
| <code>BUZZ_FORCE_CPU</code> || false || Set to <code>true</code> to disable GPU inference
|-
| <code>BUZZ_WHISPERCPP_N_THREADS</code> || cpu_count / 2 || Override thread count for whisper-cli
|}

Set via: <code>flatpak override --user io.github.chidiwilliams.Buzz --env=VAR=value</code>

== Key File Paths ==

{| class="wikitable"
|-
! Path !! Description
|-
| <code>/var/lib/flatpak/app/io.github.chidiwilliams.Buzz/.../buzz/whisper_cpp/whisper-cli</code> || Bundled Vulkan whisper-cli (libwhisper 1.8.3)
|-
| <code>~/.var/app/io.github.chidiwilliams.Buzz/cache/Buzz/models/</code> || Buzz model cache (ggml binaries)
|-
| <code>~/.var/app/io.github.chidiwilliams.Buzz/.local/state/Buzz/log/logs.txt</code> || Buzz application log
|-
| <code>~/.var/app/io.github.chidiwilliams.Buzz/config/Buzz.conf</code> || Buzz settings
|-
| <code>~/whisper.cpp/build/bin/whisper-cli</code> || Custom Vulkan-enabled whisper.cpp build
|-
| <code>~/bin/whisper-cli-vulkan</code> || Wrapper script for custom build (sets <code>LD_LIBRARY_PATH</code>)
|-
| <code>~/.local/share/pipx/venvs/buzz-captions/lib/python3.12/site-packages/buzz/whisper_cpp/</code> || pipx install whisper_cpp directory (directly modifiable)
|}

== Investigation Summary ==

{| class="wikitable"
|-
! Question !! Finding
|-
| Is Buzz using the RX 7600? || '''Yes.''' <code>using Vulkan0 backend</code> confirmed inside the Flatpak sandbox.
|-
| Was <code>--no-gpu</code> being added? || '''No.''' <code>IS_VULKAN_SUPPORTED = True</code> in the sandbox; flag is never appended.
|-
| Does the large model load correctly? || '''Yes.''' 1.6 GB to VRAM, transcribes in ~1.25s on a short clip.
|-
| What caused the crash? || '''Extract Speech (Demucs) OOM.''' Python killed at 22 GB RSS before whisper-cli launched.
|-
| Why was VRAM not climbing? || <code>whisper-cli</code> was never started — process died during Demucs pre-processing.
|-
| Fix? || '''Uncheck Extract Speech''' for long files. Clear swap before sessions.
|}

''Investigation date: 2026-04-25. System: Ubuntu 24.04, AMD RX 7600, Buzz 1.4.4 (Flatpak).''