added text cleanup step after transcription

README.md

@@ -22,5 +22,8 @@ requires `~/.config/simplewhispr/config.json`:
 }
 ```
 
+## integrations
+waybar: `tail -F /tmp/simplewhispr-waybar.log`, json output
+
 ---
 Copyright (C) Noah Swerhun under the terms of [GNU GPLv3](https://www.gnu.org/licenses/gpl-3.0-standalone.html).

main.py

@@ -10,6 +10,7 @@ from openai import OpenAI
 
 # flag to control recording
 recording = True
+status_file = None
 
 
 def parse_config() -> tuple[str, str]:
@@ -33,6 +34,21 @@ def parse_config() -> tuple[str, str]:
         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.")
@@ -56,6 +72,26 @@ def transcribe_audio(filename: str) -> str:
     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
@@ -64,6 +100,7 @@ def grab_recording() -> str:
     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
@@ -74,12 +111,14 @@ def grab_recording() -> str:
         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
 
 
@@ -87,13 +126,23 @@ 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: transcription: {transcription}")
+    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
-    subprocess.run(["wtype", transcription])
+    report_status("󰌌")
+    subprocess.run(["wtype", cleaned_transcription])
+    report_status("")
 
     if os.path.exists("/tmp/simplewhispr.pid"):
         os.remove("/tmp/simplewhispr.pid")