Add stdout option (#3027)

* add add cli options for play and speed
--play argument uses simpleaudio to play the tts wav
--speed <float 0.0-2.0> passes speed argument to Coqui Studio models

* remove simpleaudio not referenced in file

* fix simpleaudio dependency version

* add ALSA headers for simpleaudio compilation

* Dockerfile ALSA headers for simpleaudio

* base changes to use stdout instead of play audio
Considering conversion to pipe wav data for audio playback with ohter program
like aplay.

This is incomplete code. Using to get feedback before proceeding with
implementation.

* remove play for pipe_out arg that suppresses stdout
removed play and simpleaudio dependency in place of pipe
fuctionality to allow passing wav file data to a program
dedicated to playing audio.

* scipy.io.wavfile.write fails with /dev/null target

* Streaming inference for XTTS 🚀 (#3035)

* v0.17.7

* Redownload XTTS with the local and remote config do not match

* Remove unused method

* Print a message when it is already donwloaded

* Try-except to present error when the user dont have connection

* Fix style

* 0.17.8

* v0.17.8

---------

Co-authored-by: Julian Weber <julian.weber@hotmail.fr>
Co-authored-by: Eren Gölge <erogol@hotmail.com>
Co-authored-by: Edresson Casanova <edresson1@gmail.com>
Co-authored-by: ggoknar <ggoknar@coqui.ai>

README.md

@@ -347,6 +347,18 @@ If you don't specify any models, then it uses LJSpeech based English model.
   $ tts --text "Text for TTS" --out_path output/path/speech.wav
   ```
 
+- Run TTS and pipe out the generated TTS wav file data:
+
+  ```
+  $ tts --text "Text for TTS" --pipe_out --out_path output/path/speech.wav | aplay
+  ```
+
+- Run TTS and define speed factor to use for 🐸Coqui Studio models, between 0.0 and 2.0:
+
+  ```
+  $ tts --text "Text for TTS" --model_name "coqui_studio/<language>/<dataset>/<model_name>" --speed 1.2 --out_path output/path/speech.wav
+  ```
+
 - Run a TTS model with its default vocoder model:
 
   ```

TTS/api.py

@@ -112,7 +112,6 @@ class TTS(nn.Module):
             return self.synthesizer.tts_model.language_manager.num_languages > 1
         return False
 
-
     @property
     def speakers(self):
         if not self.is_multi_speaker:
@@ -265,6 +264,7 @@ class TTS(nn.Module):
         language: str = None,
         emotion: str = None,
         speed: float = 1.0,
+        pipe_out = None,
         file_path: str = None,
     ) -> Union[np.ndarray, str]:
         """Convert text to speech using Coqui Studio models. Use `CS_API` class if you are only interested in the API.
@@ -281,6 +281,8 @@ class TTS(nn.Module):
                 with "V1" model. Defaults to None.
             speed (float, optional):
                 Speed of the speech. Defaults to 1.0.
+            pipe_out (BytesIO, optional):
+                Flag to stdout the generated TTS wav file for shell pipe.
             file_path (str, optional):
                 Path to save the output file. When None it returns the `np.ndarray` of waveform. Defaults to None.
 
@@ -294,6 +296,7 @@ class TTS(nn.Module):
                 speaker_name=speaker_name,
                 language=language,
                 speed=speed,
+                pipe_out=pipe_out,
                 emotion=emotion,
                 file_path=file_path,
             )[0]
@@ -356,6 +359,7 @@ class TTS(nn.Module):
         speaker_wav: str = None,
         emotion: str = None,
         speed: float = 1.0,
+        pipe_out = None,
         file_path: str = "output.wav",
         **kwargs,
     ):
@@ -377,6 +381,8 @@ class TTS(nn.Module):
                 Emotion to use for 🐸Coqui Studio models. Defaults to "Neutral".
             speed (float, optional):
                 Speed factor to use for 🐸Coqui Studio models, between 0.0 and 2.0. Defaults to None.
+            pipe_out (BytesIO, optional):
+                Flag to stdout the generated TTS wav file for shell pipe.
             file_path (str, optional):
                 Output file path. Defaults to "output.wav".
             kwargs (dict, optional):
@@ -386,10 +392,16 @@ class TTS(nn.Module):
 
         if self.csapi is not None:
             return self.tts_coqui_studio(
-                text=text, speaker_name=speaker, language=language, emotion=emotion, speed=speed, file_path=file_path
+                text=text,
+                speaker_name=speaker,
+                language=language,
+                emotion=emotion,
+                speed=speed,
+                file_path=file_path,
+                pipe_out=pipe_out,
             )
         wav = self.tts(text=text, speaker=speaker, language=language, speaker_wav=speaker_wav, **kwargs)
-        self.synthesizer.save_wav(wav=wav, path=file_path)
+        self.synthesizer.save_wav(wav=wav, path=file_path, pipe_out=pipe_out)
         return file_path
 
     def voice_conversion(

TTS/bin/synthesize.py

@@ -2,6 +2,7 @@
 # -*- coding: utf-8 -*-
 
 import argparse
+import contextlib
 import sys
 from argparse import RawTextHelpFormatter
 
@@ -59,6 +60,18 @@ If you don't specify any models, then it uses LJSpeech based English model.
   $ tts --text "Text for TTS" --out_path output/path/speech.wav
   ```
 
+- Run TTS and pipe out the generated TTS wav file data:
+
+  ```
+  $ tts --text "Text for TTS" --pipe_out --out_path output/path/speech.wav | aplay
+  ```
+
+- Run TTS and define speed factor to use for 🐸Coqui Studio models, between 0.0 and 2.0:
+
+  ```
+  $ tts --text "Text for TTS" --model_name "coqui_studio/<language>/<dataset>/<model_name>" --speed 1.2 --out_path output/path/speech.wav
+  ```
+
 - Run a TTS model with its default vocoder model:
 
   ```
@@ -228,6 +241,20 @@ def main():
         help="Language to condition the model with. Only available for 🐸Coqui Studio `XTTS-multilingual` model.",
         default=None,
     )
+    parser.add_argument(
+        "--pipe_out",
+        help="stdout the generated TTS wav file for shell pipe.",
+        type=str2bool,
+        nargs="?",
+        const=True,
+        default=False,
+    )
+    parser.add_argument(
+        "--speed",
+        type=float,
+        help="Speed factor to use for 🐸Coqui Studio models, between 0.0 and 2.0.",
+        default=None,
+    )
 
     # args for multi-speaker synthesis
     parser.add_argument("--speakers_file_path", type=str, help="JSON file for multi-speaker model.", default=None)
@@ -335,6 +362,9 @@ def main():
     if not any(check_args):
         parser.parse_args(["-h"])
 
+    pipe_out = sys.stdout if args.pipe_out else None
+
+    with contextlib.redirect_stdout(None if args.pipe_out else sys.stdout):
         # Late-import to make things load faster
         from TTS.api import TTS
         from TTS.utils.manage import ModelManager
@@ -378,7 +408,14 @@ def main():
         if "coqui_studio" in args.model_name:
             print(" > Using 🐸Coqui Studio model: ", args.model_name)
             api = TTS(model_name=args.model_name, cs_api_model=args.cs_model)
-        api.tts_to_file(text=args.text, emotion=args.emotion, file_path=args.out_path, language=args.language)
+            api.tts_to_file(
+                text=args.text,
+                emotion=args.emotion,
+                file_path=args.out_path,
+                language=args.language,
+                speed=args.speed,
+                pipe_out=pipe_out,
+            )
             print(" > Saving output to ", args.out_path)
             return
 
@@ -495,7 +532,7 @@ def main():
 
         # save the results
         print(" > Saving output to {}".format(args.out_path))
-    synthesizer.save_wav(wav, args.out_path)
+        synthesizer.save_wav(wav, args.out_path, pipe_out=pipe_out)
 
 
 if __name__ == "__main__":

TTS/cs_api.py

@@ -9,6 +9,8 @@ import numpy as np
 import requests
 from scipy.io import wavfile
 
+from TTS.utils.audio.numpy_transforms import save_wav
+
 
 class Speaker(object):
     """Convert dict to object."""
@@ -288,6 +290,7 @@ class CS_API:
         speaker_id=None,
         emotion=None,
         speed=1.0,
+        pipe_out=None,
         language=None,
         file_path: str = None,
     ) -> str:
@@ -300,6 +303,7 @@ class CS_API:
             speaker_id (str): Speaker ID. If None, the speaker name is used.
             emotion (str): Emotion of the speaker. One of "Neutral", "Happy", "Sad", "Angry", "Dull".
             speed (float): Speed of the speech. 1.0 is normal speed.
+            pipe_out (BytesIO, optional): Flag to stdout the generated TTS wav file for shell pipe.
             language (str): Language of the text. If None, the default language of the speaker is used. Language is only
                 supported by `XTTS-multilang` model. Currently supports en, de, es, fr, it, pt, pl. Defaults to "en".
             file_path (str): Path to save the file. If None, a temporary file is created.
@@ -307,7 +311,7 @@ class CS_API:
         if file_path is None:
             file_path = tempfile.mktemp(".wav")
         wav, sr = self.tts(text, speaker_name, speaker_id, emotion, speed, language)
-        wavfile.write(file_path, sr, wav)
+        save_wav(wav=wav, path=file_path, sample_rate=sr, pipe_out=pipe_out)
         return file_path
 
 

TTS/utils/audio/numpy_transforms.py

@@ -1,3 +1,4 @@
+from io import BytesIO
 from typing import Tuple
 
 import librosa
@@ -427,16 +428,24 @@ def load_wav(*, filename: str, sample_rate: int = None, resample: bool = False,
     return x
 
 
-def save_wav(*, wav: np.ndarray, path: str, sample_rate: int = None, **kwargs) -> None:
+def save_wav(*, wav: np.ndarray, path: str, sample_rate: int = None, pipe_out = None, **kwargs) -> None:
     """Save float waveform to a file using Scipy.
 
     Args:
         wav (np.ndarray): Waveform with float values in range [-1, 1] to save.
         path (str): Path to a output file.
         sr (int, optional): Sampling rate used for saving to the file. Defaults to None.
+        pipe_out (BytesIO, optional): Flag to stdout the generated TTS wav file for shell pipe.
     """
     wav_norm = wav * (32767 / max(0.01, np.max(np.abs(wav))))
-    scipy.io.wavfile.write(path, sample_rate, wav_norm.astype(np.int16))
+
+    wav_norm = wav_norm.astype(np.int16)
+    if pipe_out:
+        wav_buffer = BytesIO()
+        scipy.io.wavfile.write(wav_buffer, sample_rate, wav_norm)
+        wav_buffer.seek(0)
+        pipe_out.buffer.write(wav_buffer.read())
+    scipy.io.wavfile.write(path, sample_rate, wav_norm)
 
 
 def mulaw_encode(*, wav: np.ndarray, mulaw_qc: int, **kwargs) -> np.ndarray:

TTS/utils/audio/processor.py

@@ -1,3 +1,4 @@
+from io import BytesIO
 from typing import Dict, Tuple
 
 import librosa
@@ -693,20 +694,27 @@ class AudioProcessor(object):
             x = self.rms_volume_norm(x, self.db_level)
         return x
 
-    def save_wav(self, wav: np.ndarray, path: str, sr: int = None) -> None:
+    def save_wav(self, wav: np.ndarray, path: str, sr: int = None, pipe_out = None) -> None:
         """Save a waveform to a file using Scipy.
 
         Args:
             wav (np.ndarray): Waveform to save.
             path (str): Path to a output file.
             sr (int, optional): Sampling rate used for saving to the file. Defaults to None.
+            pipe_out (BytesIO, optional): Flag to stdout the generated TTS wav file for shell pipe.
         """
         if self.do_rms_norm:
             wav_norm = self.rms_volume_norm(wav, self.db_level) * 32767
         else:
             wav_norm = wav * (32767 / max(0.01, np.max(np.abs(wav))))
 
-        scipy.io.wavfile.write(path, sr if sr else self.sample_rate, wav_norm.astype(np.int16))
+        wav_norm = wav_norm.astype(np.int16)
+        if pipe_out:
+            wav_buffer = BytesIO()
+            scipy.io.wavfile.write(wav_buffer, sr if sr else self.sample_rate, wav_norm)
+            wav_buffer.seek(0)
+            pipe_out.buffer.write(wav_buffer.read())
+        scipy.io.wavfile.write(path, sr if sr else self.sample_rate, wav_norm)
 
     def get_duration(self, filename: str) -> float:
         """Get the duration of a wav file using Librosa.

TTS/utils/synthesizer.py

@@ -235,19 +235,20 @@ class Synthesizer(nn.Module):
         """
         return self.seg.segment(text)
 
-    def save_wav(self, wav: List[int], path: str) -> None:
+    def save_wav(self, wav: List[int], path: str, pipe_out = None) -> None:
         """Save the waveform as a file.
 
         Args:
             wav (List[int]): waveform as a list of values.
             path (str): output path to save the waveform.
+            pipe_out (BytesIO, optional): Flag to stdout the generated TTS wav file for shell pipe.
         """
         # if tensor convert to numpy
         if torch.is_tensor(wav):
             wav = wav.cpu().numpy()
         if isinstance(wav, list):
             wav = np.array(wav)
-        save_wav(wav=wav, path=path, sample_rate=self.output_sample_rate)
+        save_wav(wav=wav, path=path, sample_rate=self.output_sample_rate, pipe_out=pipe_out)
 
     def voice_conversion(self, source_wav: str, target_wav: str) -> List[int]:
         output_wav = self.vc_model.voice_conversion(source_wav, target_wav)

tests/api_tests/test_synthesize_api.py

@@ -13,3 +13,16 @@ def test_synthesize():
         '--text "This is it" '
         f'--out_path "{output_path}"'
     )
+
+    # 🐸 Coqui studio model with speed arg.
+    run_cli(
+        'tts --model_name "coqui_studio/en/Torcull Diarmuid/coqui_studio" '
+        '--text "This is it but slow" --speed 0.1'
+        f'--out_path "{output_path}"'
+    )
+
+    # test pipe_out command
+    run_cli(
+        'tts --text "test." --pipe_out '
+        f'--out_path "{output_path}" | aplay'
+    )