Merge pull request #23 from KillianLucas/revert-22-u/shivenmian/local

Revert "fix: download whisper local model directly from huggingface"

OS/01/start.sh

@@ -1,12 +1,10 @@
 ### SETTINGS
 
 # If ALL_LOCAL is False, we'll use OpenAI's services
-# else we use whisper.cpp and piper local models
+# If setting ALL_LOCAL to true, set the path to the WHISPER local model
 export ALL_LOCAL=False
-export WHISPER_MODEL_NAME="ggml-tiny.en.bin"
-
-# Uncomment and set the OpenAI API key for OpenInterpreter to work
-# export OPENAI_API_KEY="sk-..."
+# export WHISPER_MODEL_PATH=...
+export OPENAI_API_KEY="sk-..."
 
 # Expose through Ngrok
 # Uncomment following line with your Ngrok auth token (https://dashboard.ngrok.com/get-started/your-authtoken)
@@ -40,11 +38,6 @@ export LOG_LEVEL="INFO"
 ### SETUP
 
 # 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 ${WHISPER_RUST_PATH}
-
 if [[ "$ALL_LOCAL" == "True" ]]; then
     OS=$(uname -s)
     ARCH=$(uname -m)

OS/01/stt.py

@@ -53,14 +53,13 @@ def run_command(command):
     return result.stdout, result.stderr
 
 def get_transcription_file(wav_file_path: str):
-    whisper_rust_path = os.path.join(os.path.dirname(__file__), 'local_stt', 'whisper-rust')
-    model_name = os.getenv('WHISPER_MODEL_NAME')
-    if not model_name:
-        raise EnvironmentError("WHISPER_MODEL_NAME environment variable is not set.")
+    model_path = os.getenv("WHISPER_MODEL_PATH")
+    if not model_path:
+        raise EnvironmentError("WHISPER_MODEL_PATH environment variable is not set.")
 
     output, error = run_command([
-        os.path.join(whisper_rust_path, 'whisper-rust'),
-        '--model-path', os.path.join(whisper_rust_path, model_name),
+        os.path.join(os.path.dirname(__file__), 'local_stt', 'whisper-rust', 'whisper-rust'),
+        '--model-path', model_path,
         '--file-path', wav_file_path
     ])
 

README.md

@@ -23,6 +23,8 @@ 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
@@ -30,7 +32,6 @@ 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