Merge pull request #28 from shivenmian/u/shivenmian/local

fix: auto-download STT + fixed breaking change where keyboard code unreachable

OS/01/.env.example

@@ -2,15 +2,17 @@
 # Copy this file and rename it to ".env" to use it.
 
 # If ALL_LOCAL is False, we'll use OpenAI's services
-# If setting ALL_LOCAL to true, set the path to the WHISPER local model
+# else we use whisper.cpp and piper local models
 ALL_LOCAL=False
-# WHISPER_MODEL_PATH=...
+WHISPER_MODEL_NAME="ggml-tiny.en.bin"
+
+# Uncomment and set the OpenAI API key for OpenInterpreter to work
 # OPENAI_API_KEY=sk-...
 
 # For TTS, we use the en_US-lessac-medium voice model by default
 # Please change the voice URL and voice name if you wish to use another voice
-export PIPER_VOICE_URL="https://huggingface.co/rhasspy/piper-voices/resolve/main/en/en_US/lessac/medium/"
+PIPER_VOICE_URL="https://huggingface.co/rhasspy/piper-voices/resolve/main/en/en_US/lessac/medium/"
-export PIPER_VOICE_NAME="en_US-lessac-medium.onnx"
+PIPER_VOICE_NAME="en_US-lessac-medium.onnx"
 
 # If SERVER_START, this is where we'll serve the server.
 # If DEVICE_START, this is where the device expects the server to be.

OS/01/device.py

@@ -230,10 +230,10 @@ if __name__ == "__main__":
                         toggle_recording(True)
                 else:
                     break
-            else:
+        else:
-                # Keyboard listener for spacebar press/release
+            # Keyboard listener for spacebar press/release
-                listener = keyboard.Listener(on_press=on_press, on_release=on_release)
+            listener = keyboard.Listener(on_press=on_press, on_release=on_release)
-                listener.start()
+            listener.start()
 
     asyncio.run(main())
     p.terminate()

OS/01/start.sh

@@ -9,9 +9,10 @@ set -a; source .env; set +a
 
 ### SETUP
 
-# if using local models, install the models / executables
-
 if [[ "$ALL_LOCAL" == "True" ]]; then
+    # if using local models, install the models / executables
+    WHISPER_MODEL_URL="https://huggingface.co/ggerganov/whisper.cpp/resolve/main/"
+    WHISPER_RUST_PATH="`pwd`/local_stt/whisper-rust"
     curl -OL "${WHISPER_MODEL_URL}${WHISPER_MODEL_NAME}" --output-dir ${WHISPER_RUST_PATH}
     OS=$(uname -s)
     ARCH=$(uname -m)

OS/01/stt.py

@@ -56,13 +56,14 @@ def run_command(command):
     return result.stdout, result.stderr
 
 def get_transcription_file(wav_file_path: str):
-    model_path = os.getenv("WHISPER_MODEL_PATH")
+    whisper_rust_path = os.path.join(os.path.dirname(__file__), 'local_stt', 'whisper-rust')
-    if not model_path:
+    model_name = os.getenv('WHISPER_MODEL_NAME')
-        raise EnvironmentError("WHISPER_MODEL_PATH environment variable is not set.")
+    if not model_name:
+        raise EnvironmentError("WHISPER_MODEL_NAME environment variable is not set.")
 
     output, error = run_command([
-        os.path.join(os.path.dirname(__file__), 'local_stt', 'whisper-rust', 'whisper-rust'),
+        os.path.join(whisper_rust_path, 'whisper-rust'),
-        '--model-path', model_path,
+        '--model-path', os.path.join(whisper_rust_path, model_name),
         '--file-path', wav_file_path
     ])
 

README.md

@@ -27,8 +27,6 @@ python -m pip install -r requirements.txt
 ```
 NB: Depending on your local Python version, you may run into [this issue↗](https://github.com/TaylorSMarks/playsound/issues/150) installing playsound. Workarounds are provided in the issue.
 
-If you want to run local speech-to-text from whisper, download the GGML Whisper model from [Huggingface](https://huggingface.co/ggerganov/whisper.cpp). Then in `OS/01/start.sh`, set `ALL_LOCAL=TRUE` and set `WHISPER_MODEL_PATH` to the path of the model.
-
 ## Usage
 
 ```bash
@@ -36,6 +34,7 @@ cd OS/01
 bash start.sh
 ```
 
+If you want to run local text-to-speech and speech-to-text, set `ALL_LOCAL` in the `start.sh` script to True. This will use the [whisper.cpp](https://github.com/ggerganov/whisper.cpp) and [Piper](https://github.com/rhasspy/piper) models.
 <br>
 
 ## Background