This version will use your gpu and is about 3 to 4 times faster.
Although quality is slightly lower compared to CPU.
I am getting good results (i.e less hallucinations & loops) thus with the below parameters.
If the idea is to use the transcript to have an IA turn into minutes/summary, it is plenty acceptable.
In your prompt then, mention the transcript has been generated by an IA with possible hallucinations, loops, etc and also provide a bit of context (it was a meeting about X and Z with person/function A and person/function B) : it usually provides very good results.
I have done some benchmark on a 22 minutes wave file (radio interview from 1971 with medium quality recording).
I have then asked Copilot to assess the quality.
According to whisper the GPU version + small model is the sweet spot (CPU+Small being the best).
For the record the processing time (in seconds) below.
CPU
tiny
174
CPU
base
248
CPU
small
796
GPU
tiny
26
GPU
base
43
GPU
small
120
Beware thus that this is very hardware dependant, quality wise and for now I have mixed feeling about GPU although the ratio processing time/quality is clearly (on my hardware) promoting GPU+Small.
Because at work I cannot enjoy the full copilot version and therefore cannot get transcripts (and minutes) from my teams meeting, i decided to give it a go with other tools.
I then discovered Whisper : « Whisper is an open‑source speech‑to‑text model designed by OpenAI to deliver accurate transcription across many languages. It handles real‑world audio remarkably well, even when recordings include background noise, accents, or imperfect microphone quality. Because it runs locally, it offers strong privacy and full control over your workflow. Its different model sizes—from tiny to large—let you balance speed and accuracy depending on your hardware and needs. »
Models can be downloaded here : tiny, base and small models are giving already really good results (before considering medium and large).
It is all about finding the right balance between time to process and accuracy.
I strongly recommend to also look at this GUI version here : it uses an older whisper version which is delivering at first look better results (it uses GPU) compared to the latest ggml-org whisper versions : standard, blas (preferred) or cublas (you need cuda installed and a nvidia card).
I might give it a try to openvino in a near futur (time to settle the proper python environement, script, models, etc).
In previous article, i have explained how to rclone one local folder to a remote folder.
While doing so i had enable filename encryption.
It appears that it brings nothing but complexity (content encryption is all that matters to me) and i now want to revert to original filenames while not having to reupload it all.
Here below a dry run script (to test safely) and a real run script – all this from the synology command line.
Note that we need to suffix files with .bin or else rclone will not recognize the files as encrypted.
#!/bin/bash
REMOTE_CRYPT="ovhcrypt:"
REMOTE_RAW="ovhftp:/backup/"
echo "=== DRY RUN (DECRYPT NAMES + ADD .BIN) ==="
rclone lsf -R "$REMOTE_RAW" | while read -r path; do
# Ignore les dossiers
[[ "$path" == */ ]] && continue
# Sépare dossier et fichier
dir="${path%/*}"
filename="${path##*/}"
# Filtre : noms chiffrés rclone (base32, >=16 chars)
if ! [[ "$filename" =~ ^[a-z0-9]{16,}$ ]]; then
echo "Skip (not encrypted): $path"
continue
fi
# Décrypte uniquement le nom du fichier
decfile=$(rclone cryptdecode "$REMOTE_CRYPT" "$filename" 2>/dev/null | awk '{print $NF}')
if [[ -z "$decfile" ]]; then
echo "Skip (cryptdecode failed): $path"
continue
fi
# Ajoute l'extension .bin
decfile_bin="${decfile}.bin"
# Reconstruit le chemin final
if [[ "$dir" == "$path" ]]; then
decpath="$decfile_bin"
else
decpath="$dir/$decfile_bin"
fi
echo "DRY RUN: Would rename: $path --> $decpath"
done
#!/bin/bash
REMOTE_CRYPT="ovhcrypt:"
REMOTE_RAW="ovhftp:/backup/"
echo "=== REAL RUN (DECRYPT NAMES + ADD .BIN) ==="
rclone lsf -R "$REMOTE_RAW" | while read -r path; do
# Ignore les dossiers
[[ "$path" == */ ]] && continue
# Sépare dossier et fichier
dir="${path%/*}"
filename="${path##*/}"
# Filtre : noms chiffrés rclone (base32, >=16 chars)
if ! [[ "$filename" =~ ^[a-z0-9]{16,}$ ]]; then
echo "Skip (not encrypted): $path"
continue
fi
# Décrypte uniquement le nom du fichier
decfile=$(rclone cryptdecode "$REMOTE_CRYPT" "$filename" 2>/dev/null | awk '{print $NF}')
if [[ -z "$decfile" ]]; then
echo "Skip (cryptdecode failed): $path"
continue
fi
# Ajoute l'extension .bin pour compatibilité ovhcrypt2
decfile_bin="${decfile}.bin"
# Reconstruit le chemin final
if [[ "$dir" == "$path" ]]; then
decpath="$decfile_bin"
else
decpath="$dir/$decfile_bin"
fi
echo "Renaming: $path --> $decpath"
# Renommage optimisé et OVH-friendly
rclone moveto "$REMOTE_RAW$path" "$REMOTE_RAW$decpath" \
--transfers 1 \
--checkers 1 \
--multi-thread-streams 0 \
--ftp-concurrency 1 \
--low-level-retries 3 \
--retries 3 \
--stats 10s
# Petite pause pour éviter les 421
sleep 0.2
done
Note, once all crypted filenames have been renamed to non encrypted filenames (with .bin extension), i stopped using ovhcrypt remote and switched to ovhcrypt2 remote.
my command is rclone sync /volume1/photo ovhcrypt2: –progress .
[ovhcrypt]
type = crypt
remote = ovhftp:/backup
directory_name_encryption = false
password = xxx
password2 = xxx
[ovhftp]
type = ftp
host = ftp.xxx.fr
user = username
pass = password
[ovhcrypt2]
type = crypt
remote = ovhftp:/backup
filename_encryption = off
directory_name_encryption = false
password = xxx
password2 = xxx
While on low power mode, we are changing the CPU governors to powersave (android should do that while on power saving mode but just to be sure…).
While on screen off, we are disabling cpu2 to 7 and setting cpu governors to powersave on cpu0 to 1. Also we setting lower priority on google play services. Last but not least, we are forcing doze mode to light.
This rather aggressive config paid off : i can spend a whole day on screen off without losing any battery power.