RealTimeSTT 0.3.7__tar.gz → 0.3.9__tar.gz

{realtimestt-0.3.7 → realtimestt-0.3.9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: RealtimeSTT
-Version: 0.3.7
+Version: 0.3.9
 Summary: A fast Voice Activity Detection and Transcription System
 Home-page: https://github.com/KoljaB/RealTimeSTT
 Author: Kolja Beigel
@@ -20,9 +20,9 @@ Requires-Dist: halo==0.0.31
 Requires-Dist: torch
 Requires-Dist: torchaudio
 Requires-Dist: scipy==1.14.1
+Requires-Dist: openwakeword>=0.4.0
 Requires-Dist: websockets==v12.0
 Requires-Dist: websocket-client==1.8.0
-Requires-Dist: openwakeword>=0.4.0
 Requires-Dist: numpy<2.0.0
 
 # RealtimeSTT
@@ -55,7 +55,7 @@ https://github.com/user-attachments/assets/797e6552-27cd-41b1-a7f3-e5cbc72094f5
 
 ### Updates
 
-Latest Version: v0.3.7
+Latest Version: v0.3.9
 
 See [release history](https://github.com/KoljaB/RealtimeSTT/releases).
 
@@ -129,10 +129,25 @@ pip install RealtimeSTT
 
 This will install all the necessary dependencies, including a **CPU support only** version of PyTorch.
 
-Although it is possible to run RealtimeSTT with a CPU installation only (use a small model like "tiny" or "base" in this case) you will get way better experience using:
+Although it is possible to run RealtimeSTT with a CPU installation only (use a small model like "tiny" or "base" in this case) you will get way better experience using CUDA (please scroll down).
 
-### GPU Support with CUDA (recommended)
+### Linux Installation
+
+Before installing RealtimeSTT please execute:
+
+```bash
+sudo apt-get install portaudio19-dev
+```
+
+### MacOS Installation
 
+Before installing RealtimeSTT please execute:
+
+```bash
+brew install portaudio
+```
+
+### GPU Support with CUDA (recommended)
 
 ### Updating PyTorch for CUDA Support
 
@@ -142,17 +157,17 @@ To upgrade your PyTorch installation to enable GPU support with CUDA, follow the
 To update PyTorch and Torchaudio to support CUDA 11.8, use the following commands:
 
 ```bash
-pip install torch==2.3.1+cu118 torchaudio==2.3.1 --index-url https://download.pytorch.org/whl/cu118
+pip install torch==2.5.1+cu118 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cu118
 ```
 
 #### For CUDA 12.X:
 To update PyTorch and Torchaudio to support CUDA 12.X, execute the following:
 
 ```bash
-pip install torch==2.3.1+cu121 torchaudio==2.3.1 --index-url https://download.pytorch.org/whl/cu121
+pip install torch==2.5.1+cu121 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cu121
 ```
 
-Replace `2.3.1` with the version of PyTorch that matches your system and requirements.
+Replace `2.5.1` with the version of PyTorch that matches your system and requirements.
 
 ### Steps That Might Be Necessary Before
 
@@ -462,6 +477,10 @@ When you initialize the `AudioToTextRecorder` class, you have various options to
 
 - **level** (int, default=logging.WARNING): Logging level.
 
+- **batch_size** (int, default=16): Batch size for the main transcription. Set to 0 to deactivate.
+
+- **init_logging** (bool, default=True): Whether to initialize the logging framework. Set to False to manage this yourself.
+
 - **handle_buffer_overflow** (bool, default=True): If set, the system will log a warning when an input overflow occurs during recording and remove the data from the buffer.
 
 - **beam_size** (int, default=5): The beam size to use for beam search decoding.
@@ -499,6 +518,8 @@ When you initialize the `AudioToTextRecorder` class, you have various options to
 
 - **on_realtime_transcription_stabilized**: A callback function that is triggered whenever there's an update in the real-time transcription and returns a higher quality, stabilized text as its argument.
 
+- **realtime_batch_size**: (int, default=16): Batch size for the real-time transcription model. Set to 0 to deactivate.
+
 - **beam_size_realtime** (int, default=3): The beam size to use for real-time transcription beam search decoding.
 
 #### Voice Activation Parameters

{realtimestt-0.3.7 → realtimestt-0.3.9}/README.md

@@ -28,7 +28,7 @@ https://github.com/user-attachments/assets/797e6552-27cd-41b1-a7f3-e5cbc72094f5
 
 ### Updates
 
-Latest Version: v0.3.7
+Latest Version: v0.3.9
 
 See [release history](https://github.com/KoljaB/RealtimeSTT/releases).
 
@@ -102,10 +102,25 @@ pip install RealtimeSTT
 
 This will install all the necessary dependencies, including a **CPU support only** version of PyTorch.
 
-Although it is possible to run RealtimeSTT with a CPU installation only (use a small model like "tiny" or "base" in this case) you will get way better experience using:
+Although it is possible to run RealtimeSTT with a CPU installation only (use a small model like "tiny" or "base" in this case) you will get way better experience using CUDA (please scroll down).
 
-### GPU Support with CUDA (recommended)
+### Linux Installation
+
+Before installing RealtimeSTT please execute:
+
+```bash
+sudo apt-get install portaudio19-dev
+```
+
+### MacOS Installation
+
+Before installing RealtimeSTT please execute:
 
+```bash
+brew install portaudio
+```
+
+### GPU Support with CUDA (recommended)
 
 ### Updating PyTorch for CUDA Support
 
@@ -115,17 +130,17 @@ To upgrade your PyTorch installation to enable GPU support with CUDA, follow the
 To update PyTorch and Torchaudio to support CUDA 11.8, use the following commands:
 
 ```bash
-pip install torch==2.3.1+cu118 torchaudio==2.3.1 --index-url https://download.pytorch.org/whl/cu118
+pip install torch==2.5.1+cu118 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cu118
 ```
 
 #### For CUDA 12.X:
 To update PyTorch and Torchaudio to support CUDA 12.X, execute the following:
 
 ```bash
-pip install torch==2.3.1+cu121 torchaudio==2.3.1 --index-url https://download.pytorch.org/whl/cu121
+pip install torch==2.5.1+cu121 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cu121
 ```
 
-Replace `2.3.1` with the version of PyTorch that matches your system and requirements.
+Replace `2.5.1` with the version of PyTorch that matches your system and requirements.
 
 ### Steps That Might Be Necessary Before
 
@@ -435,6 +450,10 @@ When you initialize the `AudioToTextRecorder` class, you have various options to
 
 - **level** (int, default=logging.WARNING): Logging level.
 
+- **batch_size** (int, default=16): Batch size for the main transcription. Set to 0 to deactivate.
+
+- **init_logging** (bool, default=True): Whether to initialize the logging framework. Set to False to manage this yourself.
+
 - **handle_buffer_overflow** (bool, default=True): If set, the system will log a warning when an input overflow occurs during recording and remove the data from the buffer.
 
 - **beam_size** (int, default=5): The beam size to use for beam search decoding.
@@ -472,6 +491,8 @@ When you initialize the `AudioToTextRecorder` class, you have various options to
 
 - **on_realtime_transcription_stabilized**: A callback function that is triggered whenever there's an update in the real-time transcription and returns a higher quality, stabilized text as its argument.
 
+- **realtime_batch_size**: (int, default=16): Batch size for the real-time transcription model. Set to 0 to deactivate.
+
 - **beam_size_realtime** (int, default=3): The beam size to use for real-time transcription beam search decoding.
 
 #### Voice Activation Parameters

realtimestt-0.3.9/RealtimeSTT/__init__.py

@@ -0,0 +1,3 @@
+from .audio_recorder import AudioToTextRecorder
+from .audio_recorder_client import AudioToTextRecorderClient
+from .audio_input import AudioInput

realtimestt-0.3.9/RealtimeSTT/audio_input.py

@@ -0,0 +1,153 @@
+from colorama import init, Fore, Style
+import pyaudio
+import logging
+import time
+
+
+class AudioInput:
+    def __init__(self, input_device_index=None, debug_mode=False):
+        self.input_device_index = input_device_index
+        self.debug_mode = debug_mode
+        self.audio_interface = None
+        self.stream = None
+        self.device_sample_rate = None
+
+        # PyAudio constants
+        self.CHUNK = 1024
+        self.FORMAT = pyaudio.paInt16  
+        self.CHANNELS = 1
+        self.DESIRED_RATE = 16000
+
+    def get_supported_sample_rates(self, device_index):
+        """Test which standard sample rates are supported by the specified device."""
+        standard_rates = [8000, 9600, 11025, 12000, 16000, 22050, 24000, 32000, 44100, 48000]
+        supported_rates = []
+
+        device_info = self.audio_interface.get_device_info_by_index(device_index)
+        max_channels = device_info.get('maxInputChannels')  # Changed from maxOutputChannels
+
+        for rate in standard_rates:
+            try:
+                if self.audio_interface.is_format_supported(
+                    rate,
+                    input_device=device_index,  # Changed to input_device
+                    input_channels=max_channels,  # Changed to input_channels 
+                    input_format=self.FORMAT,  # Changed to input_format
+                ):
+                    supported_rates.append(rate)
+            except:
+                continue
+        return supported_rates
+
+    def _get_best_sample_rate(self, actual_device_index, desired_rate):
+        """Determines the best available sample rate for the device."""
+        try:
+            device_info = self.audio_interface.get_device_info_by_index(actual_device_index)
+            supported_rates = self.get_supported_sample_rates(actual_device_index)
+
+            if desired_rate in supported_rates:
+                return desired_rate
+
+            return max(supported_rates)
+
+            # lower_rates = [r for r in supported_rates if r <= desired_rate]
+            # if lower_rates:
+            #     return max(lower_rates)
+
+            # higher_rates = [r for r in supported_rates if r > desired_rate]
+            # if higher_rates:
+            #     return min(higher_rates)
+
+            return int(device_info.get('defaultSampleRate', 44100))
+
+        except Exception as e:
+            logging.warning(f"Error determining sample rate: {e}")
+            return 44100  # Safe fallback
+
+    def list_devices(self):
+        """List all available audio input devices with supported sample rates."""
+        try:
+            init()  # Initialize colorama
+            self.audio_interface = pyaudio.PyAudio()
+            device_count = self.audio_interface.get_device_count()
+
+            print(f"Available audio input devices:")
+            #print(f"{Fore.LIGHTBLUE_EX}Available audio input devices:{Style.RESET_ALL}")
+            for i in range(device_count):
+                device_info = self.audio_interface.get_device_info_by_index(i)
+                device_name = device_info.get('name')
+                max_input_channels = device_info.get('maxInputChannels', 0)
+
+                if max_input_channels > 0:  # Only consider devices with input capabilities
+                    supported_rates = self.get_supported_sample_rates(i)
+                    print(f"{Fore.LIGHTGREEN_EX}Device {Style.RESET_ALL}{i}{Fore.LIGHTGREEN_EX}: {device_name}{Style.RESET_ALL}")
+                    
+                    # Format each rate in cyan
+                    if supported_rates:
+                        rates_formatted = ", ".join([f"{Fore.CYAN}{rate}{Style.RESET_ALL}" for rate in supported_rates])
+                        print(f"  {Fore.YELLOW}Supported sample rates: {rates_formatted}{Style.RESET_ALL}")
+                    else:
+                        print(f"  {Fore.YELLOW}Supported sample rates: None{Style.RESET_ALL}")
+
+        except Exception as e:
+            print(f"Error listing devices: {e}")
+        finally:
+            if self.audio_interface:
+                self.audio_interface.terminate()
+
+    def setup(self):
+        """Initialize audio interface and open stream"""
+        try:
+            self.audio_interface = pyaudio.PyAudio()
+
+            if self.debug_mode:
+                print(f"Input device index: {self.input_device_index}")
+            actual_device_index = (self.input_device_index if self.input_device_index is not None 
+                                else self.audio_interface.get_default_input_device_info()['index'])
+            
+            if self.debug_mode:
+                print(f"Actual selected device index: {actual_device_index}")
+            self.input_device_index = actual_device_index
+            self.device_sample_rate = self._get_best_sample_rate(actual_device_index, self.DESIRED_RATE)
+
+            if self.debug_mode:
+                print(f"Setting up audio on device {self.input_device_index} with sample rate {self.device_sample_rate}")
+
+            try:
+                self.stream = self.audio_interface.open(
+                    format=self.FORMAT,
+                    channels=self.CHANNELS,
+                    rate=self.device_sample_rate,
+                    input=True,
+                    frames_per_buffer=self.CHUNK,
+                    input_device_index=self.input_device_index,
+                )
+                if self.debug_mode:
+                    print(f"Audio recording initialized successfully at {self.device_sample_rate} Hz")
+                return True
+            except Exception as e:
+                print(f"Failed to initialize audio stream at {self.device_sample_rate} Hz: {e}")
+                return False
+
+        except Exception as e:
+            print(f"Error initializing audio recording: {e}")
+            if self.audio_interface:
+                self.audio_interface.terminate()
+            return False
+
+    def read_chunk(self):
+        """Read a chunk of audio data"""
+        return self.stream.read(self.CHUNK, exception_on_overflow=False)
+
+    def cleanup(self):
+        """Clean up audio resources"""
+        try:
+            if self.stream:
+                self.stream.stop_stream()
+                self.stream.close()
+                self.stream = None
+            if self.audio_interface:
+                self.audio_interface.terminate()
+                self.audio_interface = None
+        except Exception as e:
+            print(f"Error cleaning up audio resources: {e}")

{realtimestt-0.3.7 → realtimestt-0.3.9}/RealtimeSTT/audio_recorder.py

@@ -34,6 +34,7 @@ from openwakeword.model import Model
 from scipy.signal import resample
 from scipy import signal
 import signal as system_signal
+from faster_whisper import WhisperModel, BatchedInferencePipeline
 import faster_whisper
 import openwakeword
 import collections
@@ -87,11 +88,12 @@ if platform.system() != 'Darwin':
 
 
 class TranscriptionWorker:
-    def __init__(self, conn, stdout_pipe, model_path, compute_type, gpu_device_index, device,
-                 ready_event, shutdown_event, interrupt_stop_event, beam_size, initial_prompt, suppress_tokens):
+    def __init__(self, conn, stdout_pipe, model_path, download_root, compute_type, gpu_device_index, device,
+                 ready_event, shutdown_event, interrupt_stop_event, beam_size, initial_prompt, suppress_tokens, batch_size):
         self.conn = conn
         self.stdout_pipe = stdout_pipe
         self.model_path = model_path
+        self.download_root = download_root
         self.compute_type = compute_type
         self.gpu_device_index = gpu_device_index
         self.device = device
@@ -101,6 +103,7 @@ class TranscriptionWorker:
         self.beam_size = beam_size
         self.initial_prompt = initial_prompt
         self.suppress_tokens = suppress_tokens
+        self.batch_size = batch_size
         self.queue = queue.Queue()
 
     def custom_print(self, *args, **kwargs):
@@ -134,7 +137,10 @@ class TranscriptionWorker:
                 device=self.device,
                 compute_type=self.compute_type,
                 device_index=self.gpu_device_index,
+                download_root=self.download_root,
             )
+            if self.batch_size > 0:
+                model = BatchedInferencePipeline(model=model)
         except Exception as e:
             logging.exception(f"Error initializing main faster_whisper transcription model: {e}")
             raise
@@ -151,13 +157,24 @@ class TranscriptionWorker:
                 try:
                     audio, language = self.queue.get(timeout=0.1)
                     try:
-                        segments, info = model.transcribe(
-                            audio,
-                            language=language if language else None,
-                            beam_size=self.beam_size,
-                            initial_prompt=self.initial_prompt,
-                            suppress_tokens=self.suppress_tokens
-                        )
+                        if self.batch_size > 0:
+                            segments, info = model.transcribe(
+                                audio,
+                                language=language if language else None,
+                                beam_size=self.beam_size,
+                                initial_prompt=self.initial_prompt,
+                                suppress_tokens=self.suppress_tokens,
+                                batch_size=self.batch_size
+                            )
+                        else:
+                            segments, info = model.transcribe(
+                                audio,
+                                language=language if language else None,
+                                beam_size=self.beam_size,
+                                initial_prompt=self.initial_prompt,
+                                suppress_tokens=self.suppress_tokens
+                            )
+
                         transcription = " ".join(seg.text for seg in segments).strip()
                         logging.debug(f"Final text detected with main model: {transcription}")
                         self.conn.send(('success', (transcription, info)))
@@ -195,6 +212,7 @@ class AudioToTextRecorder:
 
     def __init__(self,
                  model: str = INIT_MODEL_TRANSCRIPTION,
+                 download_root: str = None, 
                  language: str = "",
                  compute_type: str = "default",
                  input_device_index: int = None,
@@ -208,6 +226,7 @@ class AudioToTextRecorder:
                  use_microphone=True,
                  spinner=True,
                  level=logging.WARNING,
+                 batch_size: int = 16,
 
                  # Realtime transcription parameters
                  enable_realtime_transcription=False,
@@ -217,6 +236,7 @@ class AudioToTextRecorder:
                  init_realtime_after_seconds=INIT_REALTIME_INITIAL_PAUSE,
                  on_realtime_transcription_update=None,
                  on_realtime_transcription_stabilized=None,
+                 realtime_batch_size: int = 16,
 
                  # Voice activation parameters
                  silero_sensitivity: float = INIT_SILERO_SENSITIVITY,
@@ -280,6 +300,8 @@ class AudioToTextRecorder:
                 'large-v2'.
                 If a specific size is provided, the model is downloaded
                 from the Hugging Face Hub.
+        - download_root (str, default=None): Specifies the root path were the Whisper models 
+          are downloaded to. When empty, the default is used. 
         - language (str, default=""): Language code for speech-to-text engine.
             If not specified, the model will attempt to detect the language
             automatically.
@@ -314,6 +336,7 @@ class AudioToTextRecorder:
         - spinner (bool, default=True): Show spinner animation with current
             state.
         - level (int, default=logging.WARNING): Logging level.
+        - batch_size (int, default=16): Batch size for the main transcription
         - enable_realtime_transcription (bool, default=False): Enables or
             disables real-time transcription of audio. When set to True, the
             audio will be transcribed continuously as it is being recorded.
@@ -345,6 +368,8 @@ class AudioToTextRecorder:
             triggered when the transcribed text stabilizes in quality. The
             stabilized text is generally more accurate but may arrive with a
             slight delay compared to the regular real-time updates.
+        - realtime_batch_size (int, default=16): Batch size for the real-time
+            transcription model.
         - silero_sensitivity (float, default=SILERO_SENSITIVITY): Sensitivity
             for the Silero Voice Activity Detection model ranging from 0
             (least sensitive) to 1 (most sensitive). Default is 0.5.
@@ -508,6 +533,7 @@ class AudioToTextRecorder:
         self.enable_realtime_transcription = enable_realtime_transcription
         self.use_main_model_for_realtime = use_main_model_for_realtime
         self.main_model_type = model
+        self.download_root = download_root
         self.realtime_model_type = realtime_model_type
         self.realtime_processing_pause = realtime_processing_pause
         self.init_realtime_after_seconds = init_realtime_after_seconds
@@ -522,6 +548,8 @@ class AudioToTextRecorder:
         self.beam_size = beam_size
         self.beam_size_realtime = beam_size_realtime
         self.allowed_latency_limit = allowed_latency_limit
+        self.batch_size = batch_size
+        self.realtime_batch_size = realtime_batch_size
 
         self.level = level
         self.audio_queue = mp.Queue()
@@ -630,7 +658,8 @@ class AudioToTextRecorder:
             args=(
                 child_transcription_pipe,
                 child_stdout_pipe,
-                model,
+                self.main_model_type,
+                self.download_root,
                 self.compute_type,
                 self.gpu_device_index,
                 self.device,
@@ -639,7 +668,8 @@ class AudioToTextRecorder:
                 self.interrupt_stop_event,
                 self.beam_size,
                 self.initial_prompt,
-                self.suppress_tokens
+                self.suppress_tokens,
+                self.batch_size
             )
         )
 
@@ -673,8 +703,11 @@ class AudioToTextRecorder:
                     model_size_or_path=self.realtime_model_type,
                     device=self.device,
                     compute_type=self.compute_type,
-                    device_index=self.gpu_device_index
+                    device_index=self.gpu_device_index,
+                    download_root=self.download_root,
                 )
+                if self.realtime_batch_size > 0:
+                    self.realtime_model_type = BatchedInferencePipeline(model=self.realtime_model_type)
 
             except Exception as e:
                 logging.exception("Error initializing faster_whisper "
@@ -2066,13 +2099,23 @@ class AudioToTextRecorder:
                                 continue
                     else:
                         # Perform transcription and assemble the text
-                        segments, info = self.realtime_model_type.transcribe(
-                            audio_array,
-                            language=self.language if self.language else None,
-                            beam_size=self.beam_size_realtime,
-                            initial_prompt=self.initial_prompt,
-                            suppress_tokens=self.suppress_tokens,
-                        )
+                        if self.realtime_batch_size > 0:
+                            segments, info = self.realtime_model_type.transcribe(
+                                audio_array,
+                                language=self.language if self.language else None,
+                                beam_size=self.beam_size_realtime,
+                                initial_prompt=self.initial_prompt,
+                                suppress_tokens=self.suppress_tokens,
+                                batch_size=self.realtime_batch_size
+                            )
+                        else:
+                            segments, info = self.realtime_model_type.transcribe(
+                                audio_array,
+                                language=self.language if self.language else None,
+                                beam_size=self.beam_size_realtime,
+                                initial_prompt=self.initial_prompt,
+                                suppress_tokens=self.suppress_tokens
+                            )
 
                         self.detected_realtime_language = info.language if info.language_probability > 0 else None
                         self.detected_realtime_language_probability = info.language_probability