pvorca 2.0.0

Orca Streaming Text-to-Speech Engine

Orca Binding for Python

Orca Streaming Text-to-Speech Engine

Made in Vancouver, Canada by Picovoice

Orca is an on-device streaming text-to-speech engine that is designed for use with LLMs, enabling zero-latency voice assistants. Orca is:

• Private; All speech synthesis runs locally.
• Cross-Platform:
  • Linux (x86_64), macOS (x86_64, arm64), Windows (x86_64, arm64)
  • Android and iOS
  • Chrome, Safari, Firefox, and Edge
  • Raspberry Pi (3, 4, 5)

Compatibility

• Python 3.9+
• Runs on Linux (x86_64), macOS (x86_64, arm64), Windows (x86_64, arm64), and Raspberry Pi (3, 4, 5).

Installation

pip3 install pvorca

AccessKey

Orca requires a valid Picovoice AccessKey at initialization. AccessKey acts as your credentials when using Orca SDKs. You can get your AccessKey for free. Make sure to keep your AccessKey secret. Signup or Login to Picovoice Console to get your AccessKey.

Usage

Orca supports two modes of operation: streaming and single synthesis. In the streaming synthesis mode, Orca processes an incoming text stream in real-time and generates audio in parallel. In the single synthesis mode, a complete text is synthesized in a single call to the Orca engine.

Create an instance of the Orca engine:

import pvorca

orca = pvorca.create(access_key='${ACCESS_KEY}')

Replace the ${ACCESS_KEY} with your AccessKey obtained from Picovoice Console.

To synthesize a text stream, create an Orca.OrcaStream object and add text to it one-by-one:

stream = orca.stream_open()

for text_chunk in text_generator():
    pcm = stream.synthesize(text_chunk)
    if pcm is not None:
        # handle pcm

pcm = stream.flush()
if pcm is not None:
    # handle pcm

The text_generator() function can be any stream generating text, for example an LLM response. Orca produces audio chunks in parallel to the incoming text stream, and returns the raw PCM whenever enough context has been added via stream.synthesize(). To ensure smooth transitions between chunks, the stream.synthesize() function returns an audio chunk that only includes the audio for a portion of the text that has been added. To generate the audio for the remaining text, stream.flush() needs to be invoked. When done with streaming text synthesis, the Orca.OrcaStream object needs to be closed:

stream.close()

If the complete text is known before synthesis, single synthesis mode can be used to generate speech in a single call to Orca:

# Return raw PCM
pcm, alignments = orca.synthesize(text='${TEXT}')

# Save the generated audio to a WAV file directly
alignments = orca.synthesize_to_file(text='${TEXT}', path='${OUTPUT_PATH}')

Replace ${TEXT} with the text to be synthesized and ${OUTPUT_PATH} with the path to save the generated audio as a single-channel 16-bit PCM WAV file. In single synthesis mode, Orca returns metadata of the synthesized audio in the form of a list of Orca.WordAlignment objects. You can print the metadata with:

for token in alignments:
    print(f"word=\"{token.word}\", start_sec={token.start_sec:.2f}, end_sec={token.end_sec:.2f}")
    for phoneme in token.phonemes:
        print(f"\tphoneme=\"{phoneme.phoneme}\", start_sec={phoneme.start_sec:.2f}, end_sec={phoneme.end_sec:.2f}")

When done make sure to explicitly release the resources using:

orca.delete()

Text input

Orca supports a wide range of English characters, including letters, numbers, symbols, and punctuation marks. You can get a list of all supported characters with the .valid_characters property. Pronunciations of characters or words not supported by this list can be achieved with custom pronunciations.

Custom pronunciations

Orca allows to embed custom pronunciations in the text via the syntax: {word|pronunciation}.
The pronunciation is expressed in ARPAbet phonemes, for example:

• "This is a {custom|K AH S T AH M} pronunciation"
• "{read|R IY D} this as {read|R EH D}, please."
• "I {live|L IH V} in {Sevilla|S EH V IY Y AH}. We have great {live|L AY V} sports!"

Language and Voice

Orca Streaming Text-to-Speech can synthesize speech in different languages and with a variety of voices, each of which is characterized by a model file (.pv) located in lib/common. The language and gender of the speaker is indicated in the file name.

To create an instance of the engine with a specific language and voice, use:

orca = pvorca.create(access_key='${ACCESS_KEY}', model_path='${MODEL_PATH}')

and replace ${MODEL_PATH} with the path to the model file with the desired language/voice.

Speech control

Orca allows for keyword arguments to control the synthesized speech. They can be provided to the stream_open method or the single synthesis methods synthesize and synthesize_to_file:

• speech_rate: Controls the speed of the generated speech. Valid values are within [0.7, 1.3]. A higher (lower) value produces speech that is faster (slower). The default is 1.0.
• random_state: Sets the random state for sampling during synthesis. This can be used to ensure that the synthesized speech is deterministic across different runs. Valid values are all non-negative integers. If not provided, a random seed will be chosen and the synthesis process will be non-deterministic.

Orca properties

To obtain the set of valid characters, call orca.valid_characters.
To retrieve the maximum number of characters allowed, call orca.max_character_limit.
The sample rate of Orca is orca.sample_rate.

Alignment Metadata

Along with the raw PCM or saved audio file, Orca returns metadata for the synthesized audio in single synthesis mode. The Orca.WordAlignment object has the following properties:

• Word: String representation of the word.
• Start Time: Indicates when the word started in the synthesized audio. Value is in seconds.
• End Time: Indicates when the word ended in the synthesized audio. Value is in seconds.
• Phonemes: A list of Orca.PhonemeAlignment objects.

The Orca.PhonemeAlignment object has the following properties:

• Phoneme: String representation of the phoneme.
• Start Time: Indicates when the phoneme started in the synthesized audio. Value is in seconds.
• End Time: Indicates when the phoneme ended in the synthesized audio. Value is in seconds.

Demos

pvorcademo provides command-line utilities for synthesizing audio using Orca.