Text To Speech Streaming With Timestamps
Converts text into audio together with timestamps on when which word was spoken in a streaming way.
Audio generation
You can generate audio together with information on when which character was spoken in a streaming way using the following script:
import requests
import json
import base64
VOICE_ID = "21m00Tcm4TlvDq8ikWAM" # Rachel
YOUR_XI_API_KEY = "ENTER_YOUR_API_KEY_HERE"
url = f"https://api.elevenlabs.io/v1/text-to-speech/{VOICE_ID}/stream/with-timestamps"
headers = {
"Content-Type": "application/json",
"xi-api-key": YOUR_XI_API_KEY
}
data = {
"text": (
"Born and raised in the charming south, "
"I can add a touch of sweet southern hospitality "
"to your audiobooks and podcasts"
),
"model_id": "eleven_multilingual_v2",
"voice_settings": {
"stability": 0.5,
"similarity_boost": 0.75
}
}
response = requests.post(
url,
json=data,
headers=headers,
stream=True
)
if response.status_code != 200:
print(f"Error encountered, status: {response.status_code}, "
f"content: {response.text}")
quit()
audio_bytes = b""
characters = []
character_start_times_seconds = []
character_end_times_seconds = []
for line in response.iter_lines():
if line: # filter out keep-alive new line
# convert the response which contains bytes into a JSON string from utf-8 encoding
json_string = line.decode("utf-8")
# parse the JSON string and load the data as a dictionary
response_dict = json.loads(json_string)
# the "audio_base64" entry in the dictionary contains the audio as a base64 encoded string,
# we need to decode it into bytes in order to save the audio as a file
audio_bytes_chunk = base64.b64decode(response_dict["audio_base64"])
audio_bytes += audio_bytes_chunk
if response_dict["alignment"] is not None:
characters.extend(response_dict["alignment"]["characters"])
character_start_times_seconds.extend(response_dict["alignment"]["character_start_times_seconds"])
character_end_times_seconds.extend(response_dict["alignment"]["character_end_times_seconds"])
with open('output.mp3', 'wb') as f:
f.write(audio_bytes)
print({
"characters": characters,
"character_start_times_seconds": character_start_times_seconds,
"character_end_times_seconds": character_end_times_seconds
})
This prints out a dictionary like:
{
'characters': ['B', 'o', 'r', 'n', ' ', 'a', 'n', 'd', ' ', 'r', 'a', 'i', 's', 'e', 'd', ' ', 'i', 'n', ' ', 't', 'h', 'e', ' ', 'c', 'h', 'a', 'r', 'm', 'i', 'n', 'g', ' ', 's', 'o', 'u', 't', 'h', ',', ' ', 'I', ' ', 'c', 'a', 'n', ' ', 'a', 'd', 'd', ' ', 'a', ' ', 't', 'o', 'u', 'c', 'h', ' ', 'o', 'f', ' ', 's', 'w', 'e', 'e', 't', ' ', 's', 'o', 'u', 't', 'h', 'e', 'r', 'n', ' ', 'h', 'o', 's', 'p', 'i', 't', 'a', 'l', 'i', 't', 'y', ' ', 't', 'o', ' ', 'y', 'o', 'u', 'r', ' ', 'a', 'u', 'd', 'i', 'o', 'b', 'o', 'o', 'k', 's', ' ', 'a', 'n', 'd', ' ', 'p', 'o', 'd', 'c', 'a', 's', 't', 's'],
'character_start_times_seconds': [0.0, 0.186, 0.279, 0.348, 0.406, 0.441, 0.476, 0.499, 0.522, 0.58, 0.65, 0.72, 0.778, 0.824, 0.882, 0.906, 0.952, 0.975, 1.01, 1.045, 1.068, 1.091, 1.115, 1.149, 1.196, 1.254, 1.3, 1.358, 1.416, 1.474, 1.498, 1.521, 1.602, 1.66, 1.811, 1.869, 1.927, 1.974, 2.009, 2.043, 2.067, 2.136, 2.183, 2.218, 2.252, 2.287, 2.322, 2.357, 2.392, 2.426, 2.45, 2.508, 2.531, 2.589, 2.635, 2.682, 2.717, 2.763, 2.786, 2.81, 2.879, 2.937, 3.007, 3.065, 3.123, 3.17, 3.239, 3.286, 3.367, 3.402, 3.437, 3.46, 3.483, 3.529, 3.564, 3.599, 3.634, 3.68, 3.75, 3.82, 3.889, 3.971, 4.087, 4.168, 4.214, 4.272, 4.331, 4.389, 4.412, 4.447, 4.528, 4.551, 4.574, 4.609, 4.644, 4.702, 4.748, 4.807, 4.865, 4.923, 5.016, 5.074, 5.12, 5.155, 5.201, 5.248, 5.283, 5.306, 5.329, 5.352, 5.41, 5.457, 5.573, 5.654, 5.735, 5.886, 5.944, 6.06],
'character_end_times_seconds': [0.186, 0.279, 0.348, 0.406, 0.441, 0.476, 0.499, 0.522, 0.58, 0.65, 0.72, 0.778, 0.824, 0.882, 0.906, 0.952, 0.975, 1.01, 1.045, 1.068, 1.091, 1.115, 1.149, 1.196, 1.254, 1.3, 1.358, 1.416, 1.474, 1.498, 1.521, 1.602, 1.66, 1.811, 1.869, 1.927, 1.974, 2.009, 2.043, 2.067, 2.136, 2.183, 2.218, 2.252, 2.287, 2.322, 2.357, 2.392, 2.426, 2.45, 2.508, 2.531, 2.589, 2.635, 2.682, 2.717, 2.763, 2.786, 2.81, 2.879, 2.937, 3.007, 3.065, 3.123, 3.17, 3.239, 3.286, 3.367, 3.402, 3.437, 3.46, 3.483, 3.529, 3.564, 3.599, 3.634, 3.68, 3.75, 3.82, 3.889, 3.971, 4.087, 4.168, 4.214, 4.272, 4.331, 4.389, 4.412, 4.447, 4.528, 4.551, 4.574, 4.609, 4.644, 4.702, 4.748, 4.807, 4.865, 4.923, 5.016, 5.074, 5.12, 5.155, 5.201, 5.248, 5.283, 5.306, 5.329, 5.352, 5.41, 5.457, 5.573, 5.654, 5.735, 5.886, 5.944, 6.06, 6.548]
}
As you can see this dictionary contains three lists of the same size. For example response_dict[‘alignment’][‘characters’][3] contains the fourth character in the text you provided ‘n’, response_dict[‘alignment’][‘character_start_times_seconds’][3] and response_dict[‘alignment’][‘character_end_times_seconds’][3] contain its start (0.348 seconds) and end (0.406 seconds) timestamps.
Headers
Your API key. This is required by most endpoints to access our API programatically. You can view your xi-api-key using the 'Profile' tab on the website.
Path Parameters
Voice ID to be used, you can use https://api.elevenlabs.io/v1/voices to list all the available voices.
Query Parameters
When enable_logging is set to false full privacy mode will be used for the request. This will mean history features are unavailable for this request, including request stitching. Full privacy mode may only be used by enterprise customers.
You can turn on latency optimizations at some cost of quality. The best possible final latency varies by model. Possible values: 0 - default mode (no latency optimizations) 1 - normal latency optimizations (about 50% of possible latency improvement of option 3) 2 - strong latency optimizations (about 75% of possible latency improvement of option 3) 3 - max latency optimizations 4 - max latency optimizations, but also with text normalizer turned off for even more latency savings (best latency, but can mispronounce eg numbers and dates).
Defaults to 0.
Output format of the generated audio. Must be one of: mp3_22050_32 - output format, mp3 with 22.05kHz sample rate at 32kbps. mp3_44100_32 - output format, mp3 with 44.1kHz sample rate at 32kbps. mp3_44100_64 - output format, mp3 with 44.1kHz sample rate at 64kbps. mp3_44100_96 - output format, mp3 with 44.1kHz sample rate at 96kbps. mp3_44100_128 - default output format, mp3 with 44.1kHz sample rate at 128kbps. mp3_44100_192 - output format, mp3 with 44.1kHz sample rate at 192kbps. Requires you to be subscribed to Creator tier or above. pcm_16000 - PCM format (S16LE) with 16kHz sample rate. pcm_22050 - PCM format (S16LE) with 22.05kHz sample rate. pcm_24000 - PCM format (S16LE) with 24kHz sample rate. pcm_44100 - PCM format (S16LE) with 44.1kHz sample rate. Requires you to be subscribed to Pro tier or above. ulaw_8000 - μ-law format (sometimes written mu-law, often approximated as u-law) with 8kHz sample rate. Note that this format is commonly used for Twilio audio inputs.
Body
The text that will get converted into speech.
Identifier of the model that will be used, you can query them using GET /v1/models. The model needs to have support for text to speech, you can check this using the can_do_text_to_speech property.
Language code (ISO 639-1) used to enforce a language for the model. Currently only Turbo v2.5 supports language enforcement. For other models, an error will be returned if language code is provided.
Voice settings overriding stored setttings for the given voice. They are applied only on the given request.
A list of pronunciation dictionary locators (id, version_id) to be applied to the text. They will be applied in order. You may have up to 3 locators per request
If specified, our system will make a best effort to sample deterministically, such that repeated requests with the same seed and parameters should return the same result. Determinism is not guaranteed.
The text that came before the text of the current request. Can be used to improve the flow of prosody when concatenating together multiple generations or to influence the prosody in the current generation.
The text that comes after the text of the current request. Can be used to improve the flow of prosody when concatenating together multiple generations or to influence the prosody in the current generation.
A list of request_id of the samples that were generated before this generation. Can be used to improve the flow of prosody when splitting up a large task into multiple requests. The results will be best when the same model is used across the generations. In case both previous_text and previous_request_ids is send, previous_text will be ignored. A maximum of 3 request_ids can be send.
A list of request_id of the samples that were generated before this generation. Can be used to improve the flow of prosody when splitting up a large task into multiple requests. The results will be best when the same model is used across the generations. In case both next_text and next_request_ids is send, next_text will be ignored. A maximum of 3 request_ids can be send.