lostecho/edge-tts
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update edge_tts to v6.1.3

Browse Source 
* Sort voices in util's --list-voices
* Disable overlapping subtitles by default
* Try to fix subtitles for many hour long TTS generation
* Match Microsoft Edge connection behavior more
* Bump edge_tts version
This commit is contained in:
rany2
2023-01-25 18:29:40 +02:00
parent 85eef7d918
commit 58307ded26
6 changed files with 56 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files

64
src/edge_tts/communicate.py
View File

@@ -254,9 +254,15 @@ class Communicate:
self.voice: str = voice
match = re.match(r"^([a-z]{2})-([A-Z]{2})-(.+Neural)$", voice)
if match is not None:
lang = match.group(1)
region = match.group(2)
name = match.group(3)
if name.find("-") != -1:
region = region + "-" + name[: name.find("-")]
name = name[name.find("-") + 1 :]
self.voice = (
"Microsoft Server Speech Text to Speech Voice"
+ f" ({match.group(1)}-{match.group(2)}, {match.group(3)})"
+ f" ({lang}-{region}, {name})"
)
if (
@@ -291,24 +297,29 @@ class Communicate:
escape(remove_incompatible_characters(self.text)),
calc_max_mesg_size(self.voice, self.rate, self.volume),
)
final_utterance: Dict[int, int] = {}
prev_idx = -1
shift_time = -1
async with aiohttp.ClientSession(trust_env=True) as session, session.ws_connect(
f"{WSS_URL}&ConnectionId={connect_id()}",
compress=15,
autoclose=True,
autoping=True,
proxy=self.proxy,
headers={
"Pragma": "no-cache",
"Cache-Control": "no-cache",
"Origin": "chrome-extension://jdiccldimpdaibmpdkjnbmckianbfold",
"Accept-Encoding": "gzip, deflate, br",
"Accept-Language": "en-US,en;q=0.9",
"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36"
" (KHTML, like Gecko) Chrome/91.0.4472.77 Safari/537.36 Edg/91.0.864.41",
},
) as websocket:
for text in texts:
for idx, text in enumerate(texts):
async with aiohttp.ClientSession(
trust_env=True
) as session, session.ws_connect(
f"{WSS_URL}&ConnectionId={connect_id()}",
compress=15,
autoclose=True,
autoping=True,
proxy=self.proxy,
headers={
"Pragma": "no-cache",
"Cache-Control": "no-cache",
"Origin": "chrome-extension://jdiccldimpdaibmpdkjnbmckianbfold",
"Accept-Encoding": "gzip, deflate, br",
"Accept-Language": "en-US,en;q=0.9",
"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36"
" (KHTML, like Gecko) Chrome/91.0.4472.77 Safari/537.36 Edg/91.0.864.41",
},
) as websocket:
# download indicates whether we should be expecting audio data,
# this is so what we avoid getting binary data from the websocket
# and falsely thinking it's audio data.
@@ -362,10 +373,25 @@ class Communicate:
elif path == b"audio.metadata":
for meta_obj in json.loads(data)["Metadata"]:
meta_type = meta_obj["Type"]
if idx != prev_idx:
shift_time = sum(
final_utterance[i] for i in range(idx)
)
prev_idx = idx
if meta_type == "WordBoundary":
final_utterance[idx] = (
meta_obj["Data"]["Offset"]
+ meta_obj["Data"]["Duration"]
# Average padding added by the service
# Alternatively we could use ffmpeg to get value properly
# but I don't want to add an additional dependency
# if this is found to work well enough.
+ 8_750_000
)
yield {
"type": meta_type,
"offset": meta_obj["Data"]["Offset"],
"offset": meta_obj["Data"]["Offset"]
+ shift_time,
"duration": meta_obj["Data"]["Duration"],
"text": meta_obj["Data"]["text"]["Text"],
}