simplewhispr/main.py

152 lines
4.1 KiB
Python

import json
import os
import signal
import subprocess
import sys
import time
from pathlib import Path
from openai import OpenAI
# flag to control recording
recording = True
status_file = None
def parse_config() -> tuple[str, str]:
config_path = Path.home() / ".config/simplewhispr/config.json"
if not config_path.exists():
raise Exception(f"fatal: config file not found at {config_path}")
try:
with open(config_path, "r") as f:
config = json.load(f)
api_key = config.get("openai_api_key")
model = config.get("model", "gpt-4o-mini-transcribe")
if not api_key:
raise ValueError("fatal: 'openai_api_key' not found in config file.")
return api_key, model
except Exception as e:
raise Exception(f"fatal: configuration error: {e}")
def report_status(message: str):
global status_file
if status_file is None:
try:
status_file = open("/tmp/simplewhispr-waybar.log", "w")
except OSError:
return
try:
status_file.write(f'{{"text":"{message}"}}\n')
status_file.flush()
except OSError:
pass
def handle_sigusr1(signum, frame):
global recording
print("\ninfo: SIGUSR1 received. stopping recording.")
recording = False
# register signal handler for SIGUSR1
signal.signal(signal.SIGUSR1, handle_sigusr1)
def transcribe_audio(filename: str) -> str:
api_key, model = parse_config()
client = OpenAI(api_key=api_key)
with open(filename, "rb") as audio_file:
transcription = client.audio.transcriptions.create(
model=model,
file=audio_file,
)
return transcription.text
def cleanup_text(text: str) -> str:
api_key, _ = parse_config()
client = OpenAI(api_key=api_key)
response = client.chat.completions.create(
model="gpt-5.4-nano",
messages=[
{
"role": "system",
"content": "IMPORTANT: your job is to clean up dictated text. you will remove filler words and correct punctuation and grammar. your goal should be to change as few of the meaningful words as possible, while removing words that are not meaningful. WARNING: do not change the phrasing or edit for clarity or style, simply remove filler words and clean up grammar.",
},
{"role": "user", "content": text},
],
)
res = response.choices[0].message.content
if res is None:
raise Exception("cleanup gave no output")
return res.strip()
def grab_recording() -> str:
output_filename = "/tmp/simplewhispr-recording.wav"
# start ffmpeg recording
cmd = ["ffmpeg", "-f", "pulse", "-i", "default", "-y", output_filename]
print(
f"info: starting recording to {output_filename}. send SIGUSR1 (kill -usr1 {os.getpid()}) to stop."
)
report_status("󰍬")
process = subprocess.Popen(
cmd, stdout=subprocess.DEVNULL, stderr=subprocess.PIPE, text=True
)
# wait for SIGUSR1
while recording:
time.sleep(0.5)
if process.poll() is not None:
print("fatal: ffmpeg exited unexpectedly.")
report_status("")
sys.exit(1)
# stop recording
process.terminate()
process.wait()
print("info: recording stopped.")
report_status("󰥔")
return output_filename
def main():
with open("/tmp/simplewhispr.pid", "w") as f:
f.write(str(os.getpid()))
report_status("󰥔")
recording_file = grab_recording()
report_status("󰙏")
print("info: transcribing...")
transcription = transcribe_audio(recording_file)
print(f"info: raw transcription: {transcription}")
print("info: cleaning up...")
cleaned_transcription = cleanup_text(transcription)
print(f"info: cleaned transcription: {cleaned_transcription}")
# use wtype to type the output
report_status("󰌌")
subprocess.run(["wtype", cleaned_transcription])
report_status("")
if os.path.exists("/tmp/simplewhispr.pid"):
os.remove("/tmp/simplewhispr.pid")
if __name__ == "__main__":
main()