File size: 1,470 Bytes
ab1b068
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
// Hindi front end in C++: Devanagari text -> phoneme symbols + nasal flags. Port of tts_hi/tiny_voca/text.py
// (normalize_text, _aksharas, _word_phones, chunks, tokenize, phonemize) and the sentence helpers of
// tts_hi/tiny_voca/infer.py (sentences, finished, shorter). Checked against Python by tools/g2p_check.cpp.
#pragma once
#include <cstdint>
#include <stdexcept>
#include <string>
#include <vector>

struct Phonemized {
    std::vector<std::string> symbols;   // IPA phones, punctuation, <bos> <eos> <sp> <unk>
    std::vector<int32_t> ids;           // indices into the model's symbol table
    std::vector<int32_t> stress;        // 1 = nasalized vowel
};

class NothingToPronounce : public std::runtime_error {
public:
    using std::runtime_error::runtime_error;
};

// symbols: the model's symbol table (GGUF tinyvoca.symbols), in id order.
class HindiG2P {
public:
    explicit HindiG2P(const std::vector<std::string>& symbols);
    Phonemized phonemize(const std::string& text) const;     // throws NothingToPronounce
private:
    std::vector<std::string> symbols_;
    int id_of(const std::string& s) const;
};

std::string normalize_text(const std::string& text);             // text.normalize_text
std::vector<std::string> split_sentences(const std::string& text);  // infer.sentences
std::string finished(const std::string& text);                   // infer.finished
std::vector<std::string> shorter(const std::string& text);       // infer.shorter