Test record
Ramble for 30 seconds. Get a clear plan for your week.
🔬 Tech tested · app test pending
Date14 Oct 2026
Tool and tierOpen-source stand-in only: faster-whisper 1.2.1 (base model, int8) on a CPU, with a synthetic voice made by gTTS 2.5.4. We have NOT tested ChatGPT voice, phone dictation, or the planning step.
Test dataFictional or public-domain only
Exact prompt
Turn what I just said into a plan: a checklist grouped by day, with deadlines. If I was vague about a date or detail, don't guess: list it under "Questions for me". Don't invent anything I didn't say.
Sample input
- Synthetic voice memo (audio) · download synthetic_memo.mp3
- What the synthetic voice says · download script.txt
Hi, it's me. Quick reminder for Saturday. We need to pick up the cake from the bakery on Elm Street before ten, call grandma to confirm she's coming, and buy candles, number seven. Oh and the party starts at two, not three.
- Messy transcript for the planning test · download messy_transcript.txt
ok so um, this week is a mess. the car needs its inspection, i think it's due end of the month? or maybe the 25th. also mum's birthday is thursday, need a gift, something garden-y. oh and the dentist called, reschedule, they said call before wednesday. and I promised Sam I'd send the photos from the trip. and pay the gym, no wait, cancel the gym actually. groceries for the weekend, people coming saturday, maybe 6 of us, someone is vegetarian.
Raw output
- Raw transcription output · download whisper_output.txt
Hi, it's me. Quick reminder for Saturday. We need to pick up the cake from the bakery on Elm Street before 10. Call grandma to confirm she's coming and buy candles. Number 7. Oh and the party starts at 2. Not 3.
What went wrong, and limits
All words were right, but punctuation changed the sense slightly: “buy candles, number seven” became “buy candles. Number 7.” Numbers came out as digits. A clean synthetic voice is easier than real speech: no noise or accent was tested. Transcription took 1.8 s for 19.2 s of audio.
Runnable test code
The script we ran (content/messy-voice-memo-plan/test_code/transcribe.py)
# Test code actually run on 2026-10-14 (box CPU, Python 3.13, faster-whisper 1.2.1, gTTS 2.5.4, ffmpeg).
# Step 1: make a synthetic voice memo (no real person's voice). Step 2: convert to 16 kHz mono WAV with ffmpeg. Step 3: transcribe.
# ffmpeg -loglevel error -y -i synthetic_memo.mp3 -ar 16000 -ac 1 memo.wav
from gtts import gTTS
t="Hi, it's me. Quick reminder for Saturday. We need to pick up the cake from the bakery on Elm Street before ten, call grandma to confirm she's coming, and buy candles, number seven. Oh and the party starts at two, not three."
gTTS(t,lang='en').save('synthetic_memo.mp3'); open('script.txt','w').write(t)
# --- after the ffmpeg conversion above ---
import time,wave,numpy as np
from faster_whisper import WhisperModel
w=wave.open('memo.wav'); a=np.frombuffer(w.readframes(w.getnframes()),np.int16).astype(np.float32)/32768
s=time.time(); m=WhisperModel("base",device="cpu",compute_type="int8")
segs,info=m.transcribe(a); out=" ".join(x.text.strip() for x in segs); open('whisper_output.txt','w').write(out)
print("SECS",round(time.time()-s,1),"AUDIO",round(len(a)/16000,1)); print(out)
Full working log
# messy-voice-memo-plan: evidence Status: TECHNICAL ONLY. faster-whisper 1.2.1 base, CPU, 2026-10-14: synthetic 19.2 s memo transcribed in 1.8 s, all words correct, numbers as digits, one punctuation change (see inputs/whisper_output.txt). Planning step and consumer apps untested. ## Test script Play inputs/synthetic_memo.mp3 (or read inputs/messy_transcript.txt aloud) into ChatGPT/Gemini voice/dictation, then the page prompt. Record transcript accuracy, plan quality, any invented dates (the transcript has vague dates: "end of month? or the 25th"; pass = flagged as a question, not fixed).