kokoro-sima / runtime /common.py
florianvoss's picture
Add minimal Kokoro Modalix runtime, CLI, webapp and speech API
dc5b70c verified
Raw History Blame Contribute Delete
5.99 kB
"""Validated web requests, sentence boundaries and bundled voice metadata."""
from __future__ import annotations
import math
from pathlib import Path
import re
ROOT = Path(__file__).resolve().parents[1]
MODELS = ROOT / "models"
LANGUAGES = {
"a": "English · American", "b": "English · British", "e": "Spanish",
"f": "French", "h": "Hindi", "i": "Italian", "j": "Japanese",
"p": "Portuguese · Brazilian", "z": "Chinese · Mandarin",
}
SAMPLES = {
"a": "The morning sunlight spilled across the kitchen table. Can you hear how naturally this voice flows? Let's try another sentence!",
"b": "Good morning! The sunlight is coming through the window. Shall we try another voice?",
"e": "¡Hola! Esta es una demostración de voz. ¿Cómo estás hoy?",
"f": "Bonjour ! Voici une démonstration de synthèse vocale. Comment allez-vous ?",
"h": "नमस्ते! यह आवाज़ का एक उदाहरण है. आप कैसे हैं?",
"i": "Ciao! Questa è una dimostrazione della voce. Come stai oggi?",
"j": "こんにちは!これは音声合成のデモです。今日はいい天気ですね。",
"p": "Olá! Esta é uma demonstração de voz. Como você está hoje?",
"z": "你好!这是语音合成演示。今天的天气怎么样?",
}
def options():
voices = []
for path in sorted((MODELS / "voices").glob("*.npy")):
name = path.stem
if len(name) < 4 or name[0] not in LANGUAGES:
continue
voices.append({"id": name, "language": name[0],
"name": name[3:].replace("_", " ").title(),
"gender": "Female" if name[1] == "f" else "Male"})
return {"languages": [{"id": k, "name": v, "sample": SAMPLES[k]} for k, v in LANGUAGES.items()],
"voices": voices, "sample_rate": 24000, "max_characters": 5000,
"defaults": {"language": "a", "voice": "af_heart", "speed": 1.0, "seed": 1234}}
def split_sentences(text):
"""Split punctuation runs, preserving decimal points and common initials."""
sentences, start, i = [], 0, 0
closing = '\"\'”’»)]}'
abbreviations = {"mr", "mrs", "ms", "dr", "prof", "sr", "jr", "st", "vs", "e.g", "i.e"}
while i < len(text):
mark = text[i]
if mark not in ".?!。?!":
i += 1
continue
if mark == ".":
if i and i + 1 < len(text) and text[i - 1].isdigit() and text[i + 1].isdigit():
i += 1
continue
word = re.search(r"([A-Za-z.]+)$", text[start:i])
value = word.group(1).lower() if word else ""
following = text[i + 1:].lstrip()
if following and (value in abbreviations or (len(value) == 1 and value.isalpha())
or (i + 2 < len(text) and text[i + 1].isalpha() and text[i + 2] == ".")):
i += 1
continue
end = i + 1
while end < len(text) and text[end] in ".?!。?!" + closing:
end += 1
sentence = text[start:end].strip()
if sentence and any(c.isalnum() for c in sentence):
sentences.append(sentence)
start, i = end, end
tail = text[start:].strip()
if tail and any(c.isalnum() for c in tail):
sentences.append(tail)
return sentences
def validate_request(data):
if not isinstance(data, dict):
raise ValueError("Expected a JSON object")
text = data.get("text", "")
if not isinstance(text, str) or not text.strip():
raise ValueError("Enter some text to generate speech.")
if len(text) > 5000:
raise ValueError("Use at most 5,000 characters per request.")
language, voice = data.get("language", "a"), data.get("voice", "af_heart")
if not isinstance(language, str) or language not in LANGUAGES:
raise ValueError("Select a supported language.")
if not isinstance(voice, str) or voice not in {v["id"] for v in options()["voices"]} or voice[0] != language:
raise ValueError("Select a voice for the chosen language.")
speed, seed = data.get("speed", 1.0), data.get("seed", 1234)
if isinstance(speed, bool) or not isinstance(speed, (int, float)) or not math.isfinite(speed) or not 0.5 <= speed <= 2.0:
raise ValueError("Speed must be between 0.5 and 2.0.")
if isinstance(seed, bool) or not isinstance(seed, int) or not 0 <= seed <= 2**32 - 1:
raise ValueError("Seed must be an integer from 0 to 4294967295.")
chunks = split_sentences(text.strip())
if not chunks:
raise ValueError("Enter text containing words or numbers.")
return {"text": text.strip(), "language": language, "voice": voice,
"speed": float(speed), "seed": seed, "sentences": chunks}
def validate_speech_request(data):
"""Supported OpenAI speech request fields; raw PCM is signed 16-bit LE."""
if not isinstance(data, dict):
raise ValueError("Expected a JSON object")
allowed = {"model", "input", "voice", "response_format", "speed", "language", "seed", "stream_format"}
unknown = set(data) - allowed
if unknown:
raise ValueError("Unsupported fields: " + ", ".join(sorted(unknown)))
if data.get("model") not in ("kokoro", "kokoro-sima"):
raise ValueError("model must be 'kokoro' or 'kokoro-sima'")
output = data.get("response_format", "wav")
if output not in ("wav", "pcm"):
raise ValueError("response_format must be 'wav' or 'pcm'")
if data.get("stream_format", "audio") != "audio":
raise ValueError("Only stream_format='audio' is supported")
voice = data.get("voice", "af_heart")
language = data.get("language", voice[:1] if isinstance(voice, str) else "a")
request = validate_request({"text": data.get("input"), "voice": voice, "language": language,
"speed": data.get("speed", 1.0), "seed": data.get("seed", 1234)})
return request, output