Text duration estimation for TTS generation.
Provides RuleDurationEstimator, which estimates audio duration from text using character phonetic weights across 600+ languages. Used by OmniVoice.generate() to determine output length when no duration is specified.
module-attribute ¶
test_cases = [
("Hindi (With complex marks)", "नमस्ते दुनिया"),
("Arabic (With vowels)", "مَرْحَبًا بِالْعَالَم"),
("Vietnamese (Lots of diacritics)", "Chào thế giới"),
("Chinese", "你好，世界！"),
("Mixed Emoji", "Hello 🌍! This is fun 🎉"),
]
instance-attribute ¶
ranges = [
(687, "latin"),
(1023, "greek"),
(1327, "cyrillic"),
(1423, "armenian"),
(1535, "hebrew"),
(1919, "arabic"),
(2207, "arabic"),
(2303, "arabic"),
(2431, "indic"),
(2559, "indic"),
(2687, "indic"),
(2815, "indic"),
(2943, "indic"),
(3071, "indic"),
(3199, "indic"),
(3327, "indic"),
(3455, "indic"),
(3583, "indic"),
(3839, "thai_lao"),
(4095, "indic"),
(4255, "khmer_myanmar"),
(4351, "georgian"),
(4607, "hangul"),
(4991, "ethiopic"),
(5023, "ethiopic"),
(5119, "default"),
(5759, "default"),
(5791, "default"),
(5887, "default"),
(5919, "default"),
(5951, "default"),
(5983, "default"),
(6015, "default"),
(6143, "khmer_myanmar"),
(6319, "default"),
(6399, "default"),
(6479, "indic"),
(6623, "indic"),
(6655, "khmer_myanmar"),
(6687, "indic"),
(6831, "indic"),
(7039, "indic"),
(7103, "indic"),
(7167, "indic"),
(7247, "indic"),
(7295, "indic"),
(7311, "cyrillic"),
(7359, "georgian"),
(7375, "indic"),
(7423, "indic"),
(7551, "latin"),
(7615, "latin"),
(7679, "default"),
(7935, "latin"),
(12447, "kana"),
(12543, "kana"),
(12591, "cjk"),
(12687, "hangul"),
(40959, "cjk"),
(42191, "yi"),
(42239, "default"),
(42559, "default"),
(42655, "cyrillic"),
(42751, "default"),
(43007, "latin"),
(43055, "indic"),
(43135, "default"),
(43231, "indic"),
(43263, "indic"),
(43311, "indic"),
(43359, "indic"),
(43391, "hangul"),
(43487, "indic"),
(43519, "khmer_myanmar"),
(43615, "indic"),
(43647, "khmer_myanmar"),
(43743, "indic"),
(43775, "indic"),
(43823, "ethiopic"),
(43887, "latin"),
(43967, "default"),
(44031, "indic"),
(55215, "hangul"),
(64255, "cjk"),
(65023, "arabic"),
(65135, "default"),
(65279, "arabic"),
(65519, "latin"),
]
instance-attribute ¶
weights = {
"cjk": 3.0,
"hangul": 2.5,
"kana": 2.2,
"ethiopic": 3.0,
"yi": 3.0,
"indic": 1.8,
"thai_lao": 1.5,
"khmer_myanmar": 1.8,
"arabic": 1.5,
"hebrew": 1.5,
"latin": 1.0,
"cyrillic": 1.0,
"greek": 1.0,
"armenian": 1.0,
"georgian": 1.0,
"punctuation": 0.5,
"space": 0.2,
"digit": 3.5,
"mark": 0.0,
"default": 1.0,
}
estimate_duration(
target_text: str,
ref_text: str,
ref_duration: float,
low_threshold: float | None = 50,
boost_strength: float = 3,
) -> float
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
target_text | str | The text for which we want to estimate the duration. | required |
ref_text | str | The reference text that was used to measure the ref_duration. | required |
ref_duration | float | The actual duration it took to speak the ref_text. | required |
low_threshold | float | The minimum duration threshold below which the estimation will be considered unreliable. | 50 |
boost_strength | float | Controls the power-curve boost for short durations. Higher values boost small durations more aggressively. 1 = no boost (linear), 2 = sqrt-like | 3 |
Returns:
| Name | Type | Description |
|---|---|---|
float | float | The estimated duration for the target_text based on the ref_text and ref_duration. |