Files
edge-tts/edge-tts.py
2021-05-23 18:10:10 +03:00

161 lines
7.8 KiB
Python
Executable File
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
#!/usr/bin/env python3
import sys
import json
import uuid
import signal
import argparse
import urllib.request
import asyncio
import ssl
import websockets
import unicodedata
from email.utils import formatdate
from xml.sax.saxutils import escape
ssl_context = ssl.create_default_context()
trustedClientToken = '6A5AA1D4EAFF4E9FB37E23D68491D6F4'
wssUrl = 'wss://speech.platform.bing.com/consumer/speech/synthesize/readaloud/edge/v1?TrustedClientToken=' + trustedClientToken
voiceList = 'https://speech.platform.bing.com/consumer/speech/synthesize/readaloud/voices/list?trustedclienttoken=' + trustedClientToken
def debug(msg, fd=sys.stderr):
if DEBUG: print(msg, file=fd)
def terminator(signo, stack_frame): sys.exit()
signal.signal(signal.SIGINT, terminator)
signal.signal(signal.SIGTERM, terminator)
def connectId(): return str(uuid.uuid4()).replace("-", "")
def removeIncompatibleControlChars(s):
output = []
for ch in s:
# We consider that these control characters are whitespace
if ch in ['\t','\n','\r']:
pass
else:
abr = unicodedata.category(ch)
if abr.startswith("C"): continue
output += [ ch ]
return "".join(output)
def list_voices():
with urllib.request.urlopen(voiceList) as url:
debug("Loading json from %s" % voiceList)
data = json.loads(url.read().decode())
debug("JSON Loaded")
for voice in data:
print()
for key in voice.keys():
debug("Processing key %s" % key)
if key in ["Name", "SuggestedCodec", "FriendlyName", "Status"]:
debug("Key %s skipped" % key)
continue
print("%s: %s" % ("Name" if key == "ShortName" else key, voice[key]))
print()
def mkssmlmsg(text=""):
message='X-RequestId:'+connectId()+'\r\nContent-Type:application/ssml+xml\r\n'
message+='X-Timestamp:'+formatdate()+'Z\r\nPath:ssml\r\n\r\n'
message+="<speak version='1.0' xmlns='http://www.w3.org/2001/10/synthesis' xml:lang='en-US'>"
message+="<voice name='" + voice + "'>" + "<prosody pitch='" + pitchString + "' rate ='" + rateString + "' volume='" + volumeString + "'>" + text + '</prosody></voice></speak>'
return message
async def run_tts(msg):
debug("Doing %s!" % msg)
async with websockets.connect(wssUrl, ssl=ssl_context) as ws:
message='X-Timestamp:'+formatdate()+'\r\nContent-Type:application/json; charset=utf-8\r\nPath:speech.config\r\n\r\n'
message+='{"context":{"synthesis":{"audio":{"metadataoptions":{"sentenceBoundaryEnabled":"'+sentenceBoundaryEnabled+'","wordBoundaryEnabled":"'+wordBoundaryEnabled+'"},"outputFormat":"' + codec + '"}}}}\r\n'
await ws.send(message)
debug("> %s" % message)
await ws.send(msg)
debug("> %s" % msg)
while True:
recv = await ws.recv()
recv = recv.encode() if type(recv) is not bytes else recv
debug("< %s" % recv)
if b'turn.end' in recv:
break
elif b'Path:audio\r\n' in recv:
sys.stdout.buffer.write(recv.split(b'Path:audio\r\n')[1])
# From https://github.com/pndurette/gTTS/blob/6d9309f05b3ad26ca356654732f3b5b9c3bec538/gtts/utils.py#L13-L54
def _minimize(the_string, delim, max_size):
"""Recursively split a string in the largest chunks
possible from the highest position of a delimiter all the way
to a maximum size
Args:
the_string (string): The string to split.
delim (string): The delimiter to split on.
max_size (int): The maximum size of a chunk.
Returns:
list: the minimized string in tokens
Every chunk size will be at minimum ``the_string[0:idx]`` where ``idx``
is the highest index of ``delim`` found in ``the_string``; and at maximum
``the_string[0:max_size]`` if no ``delim`` was found in ``the_string``.
In the latter case, the split will occur at ``the_string[max_size]``
which can be any character. The function runs itself again on the rest of
``the_string`` (``the_string[idx:]``) until no chunk is larger than
``max_size``.
"""
# Remove `delim` from start of `the_string`
# i.e. prevent a recursive infinite loop on `the_string[0:0]`
# if `the_string` starts with `delim` and is larger than `max_size`
if the_string.startswith(delim):
the_string = the_string[len(delim):]
if len(the_string) > max_size:
try:
# Find the highest index of `delim` in `the_string[0:max_size]`
# i.e. `the_string` will be cut in half on `delim` index
idx = the_string.rindex(delim, 0, max_size)
except ValueError:
# `delim` not found in `the_string`, index becomes `max_size`
# i.e. `the_string` will be cut in half arbitrarily on `max_size`
idx = max_size
# Call itself again for `the_string[idx:]`
return [the_string[:idx]] + \
_minimize(the_string[idx:], delim, max_size)
else:
return [the_string]
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Microsoft Edge's Online TTS Reader")
group = parser.add_mutually_exclusive_group(required=True)
group.add_argument('-t', '--text', help='what TTS will say')
group.add_argument('-f', '--file', help='same as --text but read from file')
parser.add_argument('-v', '--voice', help='voice for TTS. Default: en-US-AriaNeural', default='en-US-AriaNeural')
parser.add_argument('-c', '--codec', help="codec format. Default: audio-24khz-48kbitrate-mono-mp3. Another choice is webm-24khz-16bit-mono-opus", default='audio-24khz-48kbitrate-mono-mp3')
group.add_argument('-l', '--list-voices', help="lists available voices. Edge's list is incomplete so check https://bit.ly/2SFq1d3", action='store_true')
parser.add_argument('-p', '--pitch', help="set TTS pitch. Default +0Hz, For more info check https://bit.ly/3eAE5Nx", default="+0Hz")
parser.add_argument('-r', '--rate', help="set TTS rate. Default +0%%. For more info check https://bit.ly/3eAE5Nx", default="+0%")
parser.add_argument('-V', '--volume', help="set TTS volume. Default +0%%. For more info check https://bit.ly/3eAE5Nx", default="+0%")
parser.add_argument('-s', '--enable-sentence-boundary', help="enable sentence boundary (not implemented but set)", action='store_true')
parser.add_argument('-w', '--enable-word-boundary', help="enable word boundary (not implemented but set)", action='store_true')
parser.add_argument('-D', '--debug', help="some debugging", action='store_true')
args = parser.parse_args()
DEBUG = args.debug
if (args.text or args.file) is not None:
if args.file is not None:
# we need to use sys.stdin.read() because some devices
# like Windows and Termux don't have a /dev/stdin.
if args.file == "/dev/stdin":
debug("stdin detected, reading natively from stdin")
args.text = sys.stdin.read()
else:
debug("reading from %s" % args.file)
with open(args.file, 'r') as file:
args.text = file.read()
codec = args.codec
voice = args.voice
pitchString = args.pitch
rateString = args.rate
volumeString = args.volume
sentenceBoundaryEnabled = 'true' if args.enable_sentence_boundary else 'false'
wordBoundaryEnabled = 'true' if args.enable_word_boundary else 'false'
# https://hpbn.co/websocket/ says client must also send a masking key,
# which adds an extra 4 bytes to the header, resulting in 614 bytes over overhead
overhead = len(mkssmlmsg()) + 14
wsmax = 65536 - overhead
for text in _minimize(escape(removeIncompatibleControlChars(args.text)), " ", wsmax):
asyncio.get_event_loop().run_until_complete(run_tts(mkssmlmsg(text)))
elif args.list_voices:
list_voices()