💡 All-in-one AI framework for semantic search, LLM orchestration and language model workflows
Python agentsaiai-agentsembeddingsinformation-retrievallanguage-modellarge-language-modelsllmnlppythonragretrieval-augmented-generationsearchsearch-enginesemantic-searchsentence-embeddingstransformerstxtaivector-databasevector-search
Add streaming speech generation, closes #784
davidmezzettifffcdadf51e5
diff --git a/src/python/txtai/pipeline/audio/texttospeech.py b/src/python/txtai/pipeline/audio/texttospeech.pyindex cd322f3e..e85dffa0 100644--- a/src/python/txtai/pipeline/audio/texttospeech.py+++ b/src/python/txtai/pipeline/audio/texttospeech.py@@ -24,7 +24,7 @@ from ..base import Pipelineclass TextToSpeech(Pipeline):"""- Generates speech from text+ Generates speech from text."""def __init__(self, path=None, maxtokens=512):@@ -63,7 +63,7 @@ class TextToSpeech(Pipeline):# Get model input name, typically "text"self.input = self.model.get_inputs()[0].name- def __call__(self, text):+ def __call__(self, text, stream=False):"""Generates speech from text. Text longer than maxtokens will be batched and returnedas a single waveform per text input.@@ -73,6 +73,7 @@ class TextToSpeech(Pipeline):Args:text: text|list+ stream: stream response if True, defaults to FalseReturns:list of speech as NumPy array waveforms@@ -81,14 +82,12 @@ class TextToSpeech(Pipeline):# Convert results to a list if necessarytexts = [text] if isinstance(text, str) else text- outputs = []- for x in texts:- # Truncate to max size model can handle- x = self.tokenizer(x)+ # Streaming response+ if stream:+ return self.stream(texts)- # Run input through model and store result- result = self.execute(x)- outputs.append(result)+ # Transform text to speech+ outputs = [self.execute(x) for x in texts]# Return resultsreturn outputs[0] if isinstance(text, str) else outputs@@ -109,18 +108,45 @@ class TextToSpeech(Pipeline):# Default when CUDA provider isn't availablereturn ["CPUExecutionProvider"]- def execute(self, tokens):+ def stream(self, texts):"""- Executes model run for input array of tokens. This method will build batches+ Iterates over texts, splits into segments and yields snippets of audio.+ This method is designed to integrate with streaming LLM generation.++ Args:+ texts: list of input texts++ Returns:+ snippets of audio as NumPy arrays+ """++ buffer = []+ for x in texts:+ buffer.append(x)++ if x == "\n" or x.strip().endswith("."):+ data, buffer = "".join(buffer), []+ yield self.execute(data)++ if buffer:+ data = "".join(buffer)+ yield self.execute(data)++ def execute(self, text):+ """+ Executes model run for an input array of tokens. This method will build batchesof tokens when len(tokens) > maxtokens.Args:- tokens: array of tokens to pass to model+ text: text to tokenize and pass to modelReturns:waveform as NumPy array"""+ # Tokenize input+ tokens = self.tokenizer(text)+# Split into batches and processresults = []for x in self.batch(tokens, self.maxtokens):diff --git a/test/python/testpipeline/testtexttospeech.py b/test/python/testpipeline/testtexttospeech.pyindex e3fd249f..3729485b 100644--- a/test/python/testpipeline/testtexttospeech.py+++ b/test/python/testpipeline/testtexttospeech.py@@ -37,3 +37,13 @@ class TestTextToSpeech(unittest.TestCase):tts = TextToSpeech()self.assertEqual(tts.providers()[0][0], "CUDAExecutionProvider")++ def testStreaming(self):+ """+ Test streaming speech generation+ """++ tts = TextToSpeech()++ # Check that data is generated+ self.assertGreater(len(list(tts("This is a test. And another".split(), stream=True))), 0)