unity-sentis
ONNX
File size: 6,768 Bytes
23787a8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
50baaea
23787a8
 
 
50baaea
23787a8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
61a1454
23787a8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
61a1454
23787a8
 
 
73f2382
 
23787a8
 
 
 
 
 
50baaea
23787a8
50baaea
 
 
 
 
23787a8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
73f2382
23787a8
 
 
 
 
e09c786
50baaea
e09c786
50baaea
 
e09c786
50baaea
 
e09c786
23787a8
 
 
e09c786
 
23787a8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
61a1454
23787a8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
using System.Collections.Generic;
using UnityEngine;
using Unity.Sentis;
using System.IO;

//                      Jets Text-To-Speech Inference
//                      =============================
//
// This file implements the Jets Text-to-speech model in Unity Sentis
// The model uses phenomes instead of raw text so you have to convert it first.
// Place this file on the Main Camera
// Add an audio source
// Change the inputText
// When running you can press space bar to play it again

public class RunJets : MonoBehaviour
{
    public string inputText = "Once upon a time, there lived a girl called Alice. She lived in a house in the woods.";
    //string inputText = "The quick brown fox jumped over the lazy dog";
    //string inputText = "There are many uses of the things she uses!";

    //Set to true if we have put the phoneme_dict.txt in the Assets/StreamingAssets folder
    bool hasPhenomeDictionary = true;

    readonly string[] phonemes = new string[] { 
        "<blank>", "<unk>", "AH0", "N", "T", "D", "S", "R", "L", "DH", "K", "Z", "IH1", 
        "IH0", "M", "EH1", "W", "P", "AE1", "AH1", "V", "ER0", "F", ",", "AA1", "B", 
        "HH", "IY1", "UW1", "IY0", "AO1", "EY1", "AY1", ".", "OW1", "SH", "NG", "G", 
        "ER1", "CH", "JH", "Y", "AW1", "TH", "UH1", "EH2", "OW0", "EY2", "AO0", "IH2", 
        "AE2", "AY2", "AA2", "UW0", "EH0", "OY1", "EY0", "AO2", "ZH", "OW2", "AE0", "UW2", 
        "AH2", "AY0", "IY2", "AW2", "AA0", "\"", "ER2", "UH2", "?", "OY2", "!", "AW0", 
        "UH0", "OY0", "..", "<sos/eos>" };

    readonly string[] alphabet = "AE1 B K D EH1 F G HH IH1 JH K L M N AA1 P K R S T AH1 V W K Y Z".Split(' ');

    //Can change pitch and speed with this for a slightly different voice:
    const int samplerate = 22050;

    Dictionary<string, string> dict = new ();

    IWorker engine;

    AudioClip clip;

    void Start()
    {
        LoadModel();
        ReadDictionary();
        TextToSpeech();
    }

    void LoadModel()
    {
        var model = ModelLoader.Load(Path.Join(Application.streamingAssetsPath ,"jets-text-to-speech.sentis"));
        engine = WorkerFactory.CreateWorker(BackendType.GPUCompute, model);
    }

    void TextToSpeech()
    {
        string ptext;
        if (hasPhenomeDictionary)
        {
            ptext = TextToPhonemes(inputText);
            Debug.Log(ptext);
        }
        else
        {
            //If we have no phenome dictionary we can use one of these examples:
            ptext = "DH AH0 K W IH1 K B R AW1 N F AA1 K S JH AH1 M P S OW1 V ER0 DH AH0 L EY1 Z IY0 D AO1 G .";
            //ptext = "W AH1 N S AH0 P AA1 N AH0 T AY1 M , AH0 F R AA1 G M EH1 T AH0 P R IH1 N S EH0 S . DH AH0 F R AA1 G K IH1 S T DH AH0 P R IH1 N S EH0 S AH0 N D B IH0 K EY1 M AH0 P R IH1 N S .";
            //ptext = "D UW1 P L AH0 K EY2 T";
        }
        DoInference(ptext);
    }

    void ReadDictionary()
    {
        if (!hasPhenomeDictionary) return;
        string[] words = File.ReadAllLines(Path.Join(Application.streamingAssetsPath,"phoneme_dict.txt"));
        for (int i = 0; i < words.Length; i++)
        {
            string s = words[i];
            string[] parts = s.Split();
            if (parts[0] != ";;;") //ignore comments in file
            {
                string key = parts[0];
                dict.Add(key, s.Substring(key.Length + 2));
            }
        }
        // Add codes for punctuation to the dictionary
        dict.Add(",", ",");
        dict.Add(".", ".");
        dict.Add("!", "!");
        dict.Add("?", "?");
        dict.Add("\"", "\"");
        // You could add extra word pronounciations here e.g.
        //dict.Add("somenewword","[phonemes]");
    }

    public string ExpandNumbers(string text)
    {
        return text
            .Replace("0", " ZERO ")
            .Replace("1", " ONE ")
            .Replace("2", " TWO ")
            .Replace("3", " THREE ")
            .Replace("4", " FOUR ")
            .Replace("5", " FIVE ")
            .Replace("6", " SIX ")
            .Replace("7", " SEVEN ")
            .Replace("8", " EIGHT ")
            .Replace("9", " NINE ");
    }

    public string TextToPhonemes(string text)
    {
        string output = "";
        text = ExpandNumbers(text).ToUpper();

        string[] words = text.Split();
        for (int i = 0; i < words.Length; i++)
        {
            output += DecodeWord(words[i]);
        }
        return output;
    }

    //Decode the word into phenomes by looking for the longest word in the dictionary that matches
    //the first part of the word and so on. 
    //This works fairly well but could be improved. The original paper had a model that
    //dealt with guessing the phonemes of words
    public string DecodeWord(string word)
    {
        string output = "";
        int start = 0;
        for (int end = word.Length; end >= 0 && start < word.Length ; end--)
        { 
            if (end <= start) //no matches
            {
                start++;
                end = word.Length + 1;
                continue;
            }
            string subword = word.Substring(start, end - start);
            if (dict.TryGetValue(subword, out string value))
            {
                output += value + " ";
                start = end;
                end = word.Length + 1;
            }
        }
        return output;
    }
   
    int[] GetTokens(string ptext)
    {
        string[] p = ptext.Split();
        var tokens = new int[p.Length];
        for (int i = 0; i < tokens.Length; i++)
        {
            tokens[i] = Mathf.Max(0, System.Array.IndexOf(phonemes, p[i])); 
        }
        return tokens;
    }

    public void DoInference(string ptext)
    {      
        int[] tokens = GetTokens(ptext);

        using var input = new TensorInt(new TensorShape(tokens.Length), tokens);
        var result = engine.Execute(input);

        var output = result.PeekOutput("wav") as TensorFloat;
        output.CompleteOperationsAndDownload();
        var samples = output.ToReadOnlyArray();

        Debug.Log($"Audio size = {samples.Length / samplerate} seconds");

        clip = AudioClip.Create("voice audio", samples.Length, 1, samplerate, false);
        clip.SetData(samples, 0);

        Speak();
    }
    private void Speak()
    {
        AudioSource audioSource = GetComponent<AudioSource>();
        if (audioSource != null)
        {
            audioSource.clip = clip;
            audioSource.Play();
        }
        else
        {
            Debug.Log("There is no audio source");
        }
    }

    void Update()
    {
        if (Input.GetKeyDown(KeyCode.Space))
        {
            TextToSpeech();
        }
    }

    private void OnDestroy()
    {
        engine?.Dispose();
    }
}