added --api flag

AudioTranscriber.py

@@ -15,10 +15,10 @@
 MAX_PHRASES = 10
 
 class AudioTranscriber:
-    def __init__(self, mic_source, speaker_source):
+    def __init__(self, mic_source, speaker_source, model):
         self.transcript_data = {"You": [], "Speaker": []}
         self.transcript_changed_event = threading.Event()
-        self.audio_model = whisper.load_model(os.path.join(os.getcwd(), 'tiny.en.pt'))
+        self.audio_model = model
         self.audio_sources = {
             "You": {
                 "sample_rate": mic_source.SAMPLE_RATE,
@@ -46,7 +46,7 @@ def transcribe_audio_queue(self, audio_queue):
             self.update_last_sample_and_phrase_status(who_spoke, data, time_spoken)
             source_info = self.audio_sources[who_spoke]
             temp_file = source_info["process_data_func"](source_info["last_sample"])
-            text = self.get_transcription(temp_file)
+            text = self.audio_model.get_transcription(temp_file)
 
             if text != '' and text.lower() != 'you':
                 self.update_transcript(who_spoke, text, time_spoken)
@@ -81,10 +81,6 @@ def process_speaker_data(self, data):
             wf.writeframes(data)
         return temp_file
 
-    def get_transcription(self, file_path):
-        result = self.audio_model.transcribe(file_path, fp16=torch.cuda.is_available())
-        return result['text'].strip()
-
     def update_transcript(self, who_spoke, text, time_spoken):
         source_info = self.audio_sources[who_spoke]
         transcript = self.transcript_data[who_spoke]

README.md

@@ -68,7 +68,15 @@ Run the main script:
 python main.py
 ```
 
-Now, Ecoute will start transcribing your microphone input and speaker output in real-time, and provide a suggested response based on the conversation. It may take a couple of seconds to warm up before the transcription becomes real-time.
+For a better and faster version, use:
+
+```
+python main.py --api
+```
+
+Upon initiation, Ecoute will begin transcribing your microphone input and speaker output in real-time, generating a suggested response based on the conversation. Please note that it might take a few seconds for the system to warm up before the transcription becomes real-time.
+
+The --api flag significantly enhances transcription speed and accuracy, and it's expected to be the default option in future releases. However, keep in mind that using the Whisper API will consume more OpenAI credits than using the local model. This increased cost is attributed to the advanced features and capabilities that the Whisper API provides. Despite the additional cost, the considerable improvements in speed and transcription accuracy might make it a worthwhile investment for your use case.
 
 ### ⚠️ Limitations
 

TranscriberModels.py

@@ -0,0 +1,34 @@
+import openai
+import whisper
+import os
+import torch
+
+def get_model(use_api):
+    if use_api:
+        return APIWhisperTranscriber()
+    else:
+        return WhisperTranscriber()
+
+class WhisperTranscriber:
+    def __init__(self):
+        self.audio_model = whisper.load_model(os.path.join(os.getcwd(), 'tiny.en.pt'))
+        print(f"[INFO] Whisper using GPU: " + str(torch.cuda.is_available()))
+
+    def get_transcription(self, wav_file_path):
+        try:
+            result = self.audio_model.transcribe(wav_file_path, fp16=torch.cuda.is_available())
+        except Exception as e:
+            print(e)
+        return result['text'].strip()
+
+class APIWhisperTranscriber:
+    def get_transcription(self, wav_file_path):
+        new_file_path = wav_file_path + '.wav'
+        os.rename(wav_file_path, new_file_path)
+        audio_file= open(new_file_path, "rb")
+        try:
+            result = openai.Audio.translate("whisper-1", audio_file)
+        except Exception as e:
+            print(e)
+
+        return result['text'].strip()

main.py

@@ -6,6 +6,8 @@
 import queue
 import time
 import torch
+import sys
+import TranscriberModels
 
 def write_in_textbox(textbox, text):
     textbox.delete("0.0", "end")
@@ -76,15 +78,15 @@ def main():
     speaker_audio_recorder = AudioRecorder.DefaultSpeakerRecorder()
     speaker_audio_recorder.record_into_queue(audio_queue)
 
-    global_transcriber = AudioTranscriber(user_audio_recorder.source, speaker_audio_recorder.source)
-    transcribe = threading.Thread(target=global_transcriber.transcribe_audio_queue, args=(audio_queue,))
+    model = TranscriberModels.get_model('--api' in sys.argv)
+
+    transcriber = AudioTranscriber(user_audio_recorder.source, speaker_audio_recorder.source, model)
+    transcribe = threading.Thread(target=transcriber.transcribe_audio_queue, args=(audio_queue,))
     transcribe.daemon = True
     transcribe.start()
 
-    print(f"[INFO] Whisper using GPU: " + str(torch.cuda.is_available()))
-
     responder = GPTResponder()
-    respond = threading.Thread(target=responder.respond_to_transcriber, args=(global_transcriber,))
+    respond = threading.Thread(target=responder.respond_to_transcriber, args=(transcriber,))
     respond.daemon = True
     respond.start()
 
@@ -98,7 +100,7 @@ def main():
     root.grid_columnconfigure(1, weight=1)
 
      # Add the clear transcript button to the UI
-    clear_transcript_button = ctk.CTkButton(root, text="Clear Transcript", command=lambda: clear_context(global_transcriber, audio_queue, ))
+    clear_transcript_button = ctk.CTkButton(root, text="Clear Transcript", command=lambda: clear_context(transcriber, audio_queue, ))
     clear_transcript_button.grid(row=1, column=0, padx=10, pady=3, sticky="nsew")
 
     freeze_state = [False]  # Using list to be able to change its content inside inner functions
@@ -110,7 +112,7 @@ def freeze_unfreeze():
 
     update_interval_slider_label.configure(text=f"Update interval: {update_interval_slider.get()} seconds")
 
-    update_transcript_UI(global_transcriber, transcript_textbox)
+    update_transcript_UI(transcriber, transcript_textbox)
     update_response_UI(responder, response_textbox, update_interval_slider_label, update_interval_slider, freeze_state)
 
     root.mainloop()