准备工作

1.注册科大讯飞开发者账户

2.创建一个应用，获取APPID

3.下载离线sdk(你是什么平台的就下什么平台的)

4.开打压缩包

5.我们需要的文件

bin\msc.dll

bin\msc_x64.dll

bin\msc\res\tts目录的jet文件

6.创建Unity项目，把需要的plugin和tts文件依次放入工程中

了解msc.dll的函数

/// <summary>/// 引入TTSDll函数的类/// </summary>private class TTSDll{//登录，用户名，密码，appid配置[DllImport("msc_x64", CallingConvention = CallingConvention.StdCall)]public static extern int MSPLogin(string usr, string pwd, string parameters);//退出[DllImport("msc_x64", CallingConvention = CallingConvention.Winapi)]public static extern int MSPLogout();//初始化TTS会话，获取SessionID[DllImport("msc_x64", CallingConvention = CallingConvention.Winapi)]public static extern IntPtr QTTSSessionBegin(string _params, ref int errorCode);//传入需要转化的文字[DllImport("msc_x64", CallingConvention = CallingConvention.Winapi)]public static extern int QTTSTextPut(string sessionID, string textString, uint textLen, string _params);//开始获取， 注意SynthStatus[DllImport("msc_x64", CallingConvention = CallingConvention.Winapi)]public static extern IntPtr QTTSAudioGet(string sessionID, ref int audioLen, ref SynthStatus synthStatus, ref int errorCode);//结束[DllImport("msc_x64", CallingConvention = CallingConvention.Winapi)]public static extern int QTTSSessionEnd(string sessionID, string hints);//没用到[DllImport("msc_x64", CallingConvention = CallingConvention.Winapi)]public static extern int QTTSGetParam(string sessionID, string paramName, string paramValue, ref uint valueLen);}

其中的难点

通过tts文件夹得知：xiaofeng &xiaoyanSynthStatus获取状态

public enum SynthStatus{//开始获取TTS_FLAG_STILL_HAVE_DATA = 1,//获取结束TTS_FLAG_DATA_END,//取消TTS_FLAG_CMD_CANCELED}

QTTSSessionBegin的第一个参数设置

string szParams = $"engine_type = local, voice_name = { m_Speeker }, text_encoding = UTF8, tts_res_path = { TTSPath(m_Speeker) }, sample_rate = 16000, speed = { m_Speed }, volume = { m_Vol }, pitch = 50, rdn = 2";

离线语音合成 Windows SDK 文档 | 讯飞开放平台文档中心/doc/tts/offline_tts/Windows-SDK.html#_2-3-%E5%8F%82%E6%95%B0%E4%B8%8E%E8%AF%B4%E6%98%8E

sdk默认生成wav格式的音频，但是流文件保存的wav文件无法播放怎么办？

获取的流文件如何在unity中使用AudioSource播出？

生成正确的WAV文件

我们先看下sdk的例子

/** 语音合成（Text To Speech，TTS）技术能够自动将任意文字实时转换为连续的* 自然语音，是一种能够在任何时间、任何地点，向任何人提供语音信息服务的* 高效便捷手段，非常符合信息时代海量数据、动态更新和个性化查询的需求。*/#include <stdlib.h>#include <stdio.h>#include <windows.h>#include <conio.h>#include <errno.h>#include "qtts.h"#include "msp_cmn.h"#include "msp_errors.h"#ifdef _WIN64#pragma comment(lib,"../../libs/msc_x64.lib")//x64#else#pragma comment(lib,"../../libs/msc.lib")//x86#endif/* wav音频头部格式 */typedef struct _wave_pcm_hdr{char riff[4];// = "RIFF"intsize_8; // = FileSize - 8char wave[4];// = "WAVE"char fmt[4]; // = "fmt "intfmt_size;// = 下一个结构体的大小 : 16short int format_tag; // = PCM : 1short int channels;// = 通道数 : 1intsamples_per_sec; // = 采样率 : 8000 | 6000 | 11025 | 16000intavg_bytes_per_sec;// = 每秒字节数 : samples_per_sec * bits_per_sample / 8short int block_align; // = 每采样点字节数 : wBitsPerSample / 8short int bits_per_sample; // = 量化比特数: 8 | 16char data[4];// = "data";intdata_size; // = 纯数据长度 : FileSize - 44 } wave_pcm_hdr;/* 默认wav音频头部数据 */wave_pcm_hdr default_wav_hdr = {{ 'R', 'I', 'F', 'F' },0,{'W', 'A', 'V', 'E'},{'f', 'm', 't', ' '},16,1,1,16000,32000,2,16,{'d', 'a', 't', 'a'},0 };/* 文本合成 */int text_to_speech(const char* src_text, const char* des_path, const char* params){intret= -1;FILE* fp = NULL;const char* sessionID = NULL;unsigned int audio_len = 0;wave_pcm_hdr wav_hdr= default_wav_hdr;intsynth_status = MSP_TTS_FLAG_STILL_HAVE_DATA;if (NULL == src_text || NULL == des_path){printf("params is error!\n");return ret;}fp = fopen(des_path, "wb");if (NULL == fp){printf("open %s error.\n", des_path);return ret;}/* 开始合成 */sessionID = QTTSSessionBegin(params, &ret);if (MSP_SUCCESS != ret){printf("QTTSSessionBegin failed, error code: %d.\n", ret);fclose(fp);return ret;}ret = QTTSTextPut(sessionID, src_text, (unsigned int)strlen(src_text), NULL);if (MSP_SUCCESS != ret){printf("QTTSTextPut failed, error code: %d.\n",ret);QTTSSessionEnd(sessionID, "TextPutError");fclose(fp);return ret;}printf("正在合成 ...\n");fwrite(&wav_hdr, sizeof(wav_hdr) ,1, fp); //添加wav音频头，使用采样率为16000while (1) {/* 获取合成音频 */const void* data = QTTSAudioGet(sessionID, &audio_len, &synth_status, &ret);if (MSP_SUCCESS != ret)break;if (NULL != data){fwrite(data, audio_len, 1, fp);wav_hdr.data_size += audio_len; //计算data_size大小}if (MSP_TTS_FLAG_DATA_END == synth_status)break;}printf("\n");if (MSP_SUCCESS != ret){printf("QTTSAudioGet failed, error code: %d.\n",ret);QTTSSessionEnd(sessionID, "AudioGetError");fclose(fp);return ret;}/* 修正wav文件头数据的大小 */wav_hdr.size_8 += wav_hdr.data_size + (sizeof(wav_hdr) - 8);/* 将修正过的数据写回文件头部,音频文件为wav格式 */fseek(fp, 4, 0);fwrite(&wav_hdr.size_8,sizeof(wav_hdr.size_8), 1, fp); //写入size_8的值fseek(fp, 40, 0); //将文件指针偏移到存储data_size值的位置fwrite(&wav_hdr.data_size,sizeof(wav_hdr.data_size), 1, fp); //写入data_size的值fclose(fp);fp = NULL;/* 合成完毕 */ret = QTTSSessionEnd(sessionID, "Normal");if (MSP_SUCCESS != ret){printf("QTTSSessionEnd failed, error code: %d.\n",ret);}return ret;}int main(int argc, char* argv[]){int ret = MSP_SUCCESS;const char* login_params = "appid = XXXXXXXX, work_dir = .";//登录参数,appid与msc库绑定,请勿随意改动/** rdn: 合成音频数字发音方式* volume: 合成音频的音量* pitch: 合成音频的音调* speed: 合成音频对应的语速* voice_name: 合成发音人* sample_rate: 合成音频采样率* text_encoding: 合成文本编码格式**/const char* session_begin_params = "engine_type = local, voice_name = xiaoyan, text_encoding = GB2312, tts_res_path = fo|res\\tts\\xiaoyan.jet;fo|res\\tts\\common.jet, sample_rate = 16000, speed = 50, volume = 50, pitch = 50, rdn = 2";const char* filename = "tts_sample.wav"; //合成的语音文件名称const char* text = "科大讯飞作为智能语音技术提供商，在智能语音技术领域有着长期的研究积累，并在中文语音合成、语音识别、口语评测等多项技术上拥有技术成果。科大讯飞是我国以语音技术为产业化方向的国家863计划产业化基地"; //合成文本/* 用户登录 */ret = MSPLogin(NULL, NULL, login_params); //第一个参数是用户名，第二个参数是密码，第三个参数是登录参数，用户名和密码可在注册获取if (MSP_SUCCESS != ret){printf("MSPLogin failed, error code: %d.\n", ret);goto exit ;//登录失败，退出登录}printf("\n###########################################################################\n");printf("## 语音合成（Text To Speech，TTS）技术能够自动将任意文字实时转换为连续的 ##\n");printf("## 自然语音，是一种能够在任何时间、任何地点，向任何人提供语音信息服务的 ##\n");printf("## 高效便捷手段，非常符合信息时代海量数据、动态更新和个性化查询的需求。 ##\n");printf("###########################################################################\n\n");/* 文本合成 */printf("开始合成 ...\n");ret = text_to_speech(text, filename, session_begin_params);if (MSP_SUCCESS != ret){printf("text_to_speech failed, error code: %d.\n", ret);}printf("合成完毕\n");exit:printf("按任意键退出 ...\n");_getch();MSPLogout(); //退出登录return 0;}

看完后气不气，还特么需要手动添加wav的文件头，这样生成的wav文件才能播放

那么我们就针对这个编写C#的wav文件头

using System;using System.Collections;using System.Collections.Generic;using System.Runtime.InteropServices;using System.Text;using UnityEngine;/// <summary>/// wave文件头/// </summary>public struct WAVE_Header{public int RIFF_ID; //4 byte , 'RIFF'public int File_Size; //4 byte , 文件长度public int RIFF_Type; //4 byte , 'WAVE'public int FMT_ID; //4 byte , 'fmt'public int FMT_Size;//4 byte , 数值为16或18，18则最后又附加信息public short FMT_Tag; //2 byte , 编码方式，一般为0x0001public ushort FMT_Channel; //2 byte , 声道数目，1--单声道；2--双声道public int FMT_SamplesPerSec; //4 byte , 采样频率public int AvgBytesPerSec; //4 byte , 每秒所需字节数,记录每秒的数据量public ushort BlockAlign;//2 byte , 数据块对齐单位(每个采样需要的字节数)public ushort BitsPerSample; //2 byte , 每个采样需要的bit数public int DATA_ID; //4 byte , 'data'public int DATA_Size; //4 byte , }public class ToWave{/// <summary>/// 根据数据段的长度，生产文件头/// </summary>/// <param name="data_len">音频数据长度</param>/// <returns>返回wav文件头结构体</returns>public static WAVE_Header getWave_Header(int data_len){WAVE_Header wav_Header = new WAVE_Header();wav_Header.RIFF_ID = 0x46464952; //字符RIFFwav_Header.File_Size = data_len + 36;wav_Header.RIFF_Type = 0x45564157;//字符WAVEwav_Header.FMT_ID = 0x20746D66; //字符fmtwav_Header.FMT_Size = 16;wav_Header.FMT_Tag = 0x0001;wav_Header.FMT_Channel = 1; //单声道wav_Header.FMT_SamplesPerSec = 16000; //采样频率wav_Header.AvgBytesPerSec = 32000;//每秒所需字节数wav_Header.BlockAlign = 2; //每个采样1个字节wav_Header.BitsPerSample = 16; //每个采样8bitwav_Header.DATA_ID = 0x61746164; //字符datawav_Header.DATA_Size = data_len;return wav_Header;}/// <summary>/// 把结构体转化为字节序列/// </summary>/// <param name="structure">被转化的结构体</param>/// <returns>返回字节序列</returns>public static byte[] StructToBytes(object structure){int size = Marshal.SizeOf(structure);IntPtr buffer = Marshal.AllocHGlobal(size);try{Marshal.StructureToPtr(structure, buffer, false);Byte[] bytes = new Byte[size];Marshal.Copy(buffer, bytes, 0, size);return bytes;}finally{Marshal.FreeHGlobal(buffer);}}}

结合示例编写C#的离线语音合成

public class XunFeiTemplate{private string m_SessionID;private string m_Speed;private string m_Vol;private string m_Speeker;private Dictionary<Speeker, string> DSpeeker = new Dictionary<Speeker, string>();/// <summary>/// 构造函数，初始化引擎/// </summary>/// <param name="configs">初始化引擎参数</param>/// <param name="szParams">开始会话用参数</param>public XunFeiTemplate(string name, string password, string configs){DSpeeker.Add(Speeker.XiaoYan_Woman_Mandarin, "xiaoyan");DSpeeker.Add(Speeker.XiaoFeng_Man_Mandarin, "xiaofeng");int ret = TTSDll.MSPLogin(name, password, configs);if (ret != 0) throw new Exception("初始化TTS引擎错误，错误代码：" + ret);m_Speed = "50";m_Vol = "100";SetSpeaker(Speeker.XiaoYan_Woman_Mandarin);UnityEngine.Debug.Log(TTSPath(m_Speeker));}public void SetSpeaker(Speeker speeker){if (DSpeeker.ContainsKey(speeker))m_Speeker = DSpeeker[speeker];}public void CloseXunFei(){int ret = TTSDll.MSPLogout();if (ret != 0) throw new Exception("逆初始化TTS引擎错误，错误代码：" + ret);}public string TTSPath(string speeker){//fo|C:\\Users\\Administrator\\Desktop\\tts\\xiaofeng.jet;fo|C:\\Users\\Administrator\\Desktop\\tts\\common.jetstring path = UnityEngine.Application.streamingAssetsPath;path = path.Replace("/", "\\");string combine = "fo|{0}\\tts\\{1}.jet;fo|{2}\\tts\\common.jet";return string.Format(combine, path, speeker, path);}/// <summary>/// 把文本转换成声音，写入指定的内存流/// </summary>/// <param name="speekText">要转化成语音的文字</param>/// <param name="mStream">合成结果输出的音频流</param>private void Speek(string speekText, ref MemoryStream mStream){if (speekText == "" || m_Speed == "" || m_Vol == "" || m_Speeker == "") return;string szParams = $"engine_type = local, voice_name = { m_Speeker }, text_encoding = UTF8, tts_res_path = { TTSPath(m_Speeker) }, sample_rate = 16000, speed = { m_Speed }, volume = { m_Vol }, pitch = 50, rdn = 2";int ret = 0;try{m_SessionID = Marshal.PtrToStringAnsi(TTSDll.QTTSSessionBegin(szParams, ref ret));if (ret != 0){UnityEngine.Debug.Log(ret);throw new Exception("初始化TTS会话错误，错误代码：" + ret);}ret = TTSDll.QTTSTextPut(m_SessionID, speekText, (uint)Encoding.Default.GetByteCount(speekText), null);if (ret != 0){UnityEngine.Debug.Log(ret);throw new Exception("发送数据，错误代码：" + ret);}SynthStatus synth_status = SynthStatus.TTS_FLAG_STILL_HAVE_DATA;while (true){int audio_len = 0;IntPtr source = TTSDll.QTTSAudioGet(m_SessionID, ref audio_len, ref synth_status, ref ret);byte[] array = new byte[audio_len];if (audio_len > 0){Marshal.Copy(source, array, 0, audio_len);}mStream.Write(array, 0, audio_len);//将合成的音频字节数据存放到内存流中//Thread.Sleep(15);//防止CPU频繁占用if (synth_status == SynthStatus.TTS_FLAG_DATA_END || ret != 0)break;}}catch (Exception ex){UnityEngine.Debug.Log(ex.Message);}finally{ret = TTSDll.QTTSSessionEnd(m_SessionID, "");if (ret != 0) throw new Exception("结束TTS会话错误，错误代码：" + ret);}}public MemoryStream SpeechSynthesis(string SpeekText){MemoryStream mStream = new MemoryStream(1024 * 8);mStream.Write(new byte[44], 0, 44);Speek(SpeekText, ref mStream);//创建wav文件头WAVE_Header header = ToWave.getWave_Header((int)mStream.Length - 44);//把文件头结构转化为字节数组 byte[] headerByte = ToWave.StructToBytes(header);//定位到文件头mStream.Position = 0;//写入文件头mStream.Write(headerByte, 0, headerByte.Length);return mStream;}}

最终获得的MemoryStream就能保存为正常的wav文件了

MemoryStream mStream = new MemoryStream(1024 * 8);Speek(SpeekText, ref mStream);using (FileStream wavFile = new FileStream("music/test.wav", FileMode.Create)){mStream.WriteTo(wavFile);}

PS：TTSDll.MSPLogin(name, password, configs);

configs的配置按示例程序中

用户名和密码可以传null

TTSPath获得的是绝对路径，sdk里面获取的是相对路径，unity用不了所以使用绝对路径

在Unity中播放流媒体

当然，你也可以保存为wav文件，通过动态加载的方式，让AudioSource播放。

但是，我们既然获得了流媒体，为什么不直接在AudioSource播放呢？

AudioClip底层有个SetData的方法，那么就证明这个想法是可行的，我们只要把二进制流转为AudioClip中的数据就可以了。

wav格式解析自行百度理解。

public boolSetData(float[]data, intoffsetSamples);

设置剪辑的采样数据。

采样的浮点数范围是-1.0 ~ 1.0（超过这个限制会引发瑕疵或未知的行为），采样数由浮点数数组的长度确定。使用offsetSamples从剪辑的随机位置开始读。如果从偏移值读取的长度比剪辑长，读取将环绕从剪辑开始读取剩余采样。

PS：仅支持未压缩的PCM wav文件。

public static AudioClip Clip(byte[] fileBytes, int offsetSamples = 0, string name = "ifly"){//string riff = Encoding.ASCII.GetString (fileBytes, 0, 4);//string wave = Encoding.ASCII.GetString (fileBytes, 8, 4);int subchunk1 = BitConverter.ToInt32(fileBytes, 16);ushort audioFormat = BitConverter.ToUInt16(fileBytes, 20);// NB: Only uncompressed PCM wav files are supported.string formatCode = FormatCode(audioFormat);//Debug.AssertFormat(audioFormat == 1 || audioFormat == 65534, "Detected format code '{0}' {1}, but only PCM and WaveFormatExtensable uncompressed formats are currently supported.", audioFormat, formatCode);ushort channels = BitConverter.ToUInt16(fileBytes, 22);int sampleRate = BitConverter.ToInt32(fileBytes, 24);//int byteRate = BitConverter.ToInt32 (fileBytes, 28);//UInt16 blockAlign = BitConverter.ToUInt16 (fileBytes, 32);ushort bitDepth = BitConverter.ToUInt16(fileBytes, 34);int headerOffset = 16 + 4 + subchunk1 + 4;int subchunk2 = BitConverter.ToInt32(fileBytes, headerOffset);//Debug.LogFormat ("riff={0} wave={1} subchunk1={2} format={3} channels={4} sampleRate={5} byteRate={6} blockAlign={7} bitDepth={8} headerOffset={9} subchunk2={10} filesize={11}", riff, wave, subchunk1, formatCode, channels, sampleRate, byteRate, blockAlign, bitDepth, headerOffset, subchunk2, fileBytes.Length);//Log.Info(bitDepth);float[] data;switch (bitDepth){case 8:data = Convert8BitByteArrayToAudioClipData(fileBytes, headerOffset, subchunk2);break;case 16:data = Convert16BitByteArrayToAudioClipData(fileBytes, headerOffset, subchunk2);break;case 24:data = Convert24BitByteArrayToAudioClipData(fileBytes, headerOffset, subchunk2);break;case 32:data = Convert32BitByteArrayToAudioClipData(fileBytes, headerOffset, subchunk2);break;default:throw new Exception(bitDepth + " bit depth is not supported.");}AudioClip audioClip = AudioClip.Create(name, data.Length, channels, sampleRate, false);audioClip.SetData(data, 0);return audioClip;}private static string FormatCode(UInt16 code){switch (code){case 1:return "PCM";case 2:return "ADPCM";case 3:return "IEEE";case 7:return "μ-law";case 65534:return "WaveFormatExtensable";default:Debug.LogWarning("Unknown wav code format:" + code);return "";}}#region wav file bytes to Unity AudioClip conversion methodsprivate static float[] Convert8BitByteArrayToAudioClipData(byte[] source, int headerOffset, int dataSize){int wavSize = BitConverter.ToInt32(source, headerOffset);headerOffset += sizeof(int);Debug.AssertFormat(wavSize > 0 && wavSize == dataSize, "Failed to get valid 8-bit wav size: {0} from data bytes: {1} at offset: {2}", wavSize, dataSize, headerOffset);float[] data = new float[wavSize];sbyte maxValue = sbyte.MaxValue;int i = 0;while (i < wavSize){data[i] = (float)source[i] / maxValue;++i;}return data;}private static float[] Convert16BitByteArrayToAudioClipData(byte[] source, int headerOffset, int dataSize){int wavSize = BitConverter.ToInt32(source, headerOffset);headerOffset += sizeof(int);Debug.AssertFormat(wavSize > 0 && wavSize == dataSize, "Failed to get valid 16-bit wav size: {0} from data bytes: {1} at offset: {2}", wavSize, dataSize, headerOffset);int x = sizeof(Int16); // block size = 2int convertedSize = wavSize / x;float[] data = new float[convertedSize];Int16 maxValue = Int16.MaxValue;int offset = 0;int i = 0;while (i < convertedSize){offset = i * x + headerOffset;data[i] = (float)BitConverter.ToInt16(source, offset) / maxValue;++i;}Debug.AssertFormat(data.Length == convertedSize, "AudioClip .wav data is wrong size: {0} == {1}", data.Length, convertedSize);return data;}private static float[] Convert24BitByteArrayToAudioClipData(byte[] source, int headerOffset, int dataSize){int wavSize = BitConverter.ToInt32(source, headerOffset);headerOffset += sizeof(int);Debug.AssertFormat(wavSize > 0 && wavSize == dataSize, "Failed to get valid 24-bit wav size: {0} from data bytes: {1} at offset: {2}", wavSize, dataSize, headerOffset);int x = 3; // block size = 3int convertedSize = wavSize / x;int maxValue = Int32.MaxValue;float[] data = new float[convertedSize];byte[] block = new byte[sizeof(int)]; // using a 4 byte block for copying 3 bytes, then copy bytes with 1 offsetint offset = 0;int i = 0;while (i < convertedSize){offset = i * x + headerOffset;Buffer.BlockCopy(source, offset, block, 1, x);data[i] = (float)BitConverter.ToInt32(block, 0) / maxValue;++i;}Debug.AssertFormat(data.Length == convertedSize, "AudioClip .wav data is wrong size: {0} == {1}", data.Length, convertedSize);return data;}private static float[] Convert32BitByteArrayToAudioClipData(byte[] source, int headerOffset, int dataSize){int wavSize = BitConverter.ToInt32(source, headerOffset);headerOffset += sizeof(int);Debug.AssertFormat(wavSize > 0 && wavSize == dataSize, "Failed to get valid 32-bit wav size: {0} from data bytes: {1} at offset: {2}", wavSize, dataSize, headerOffset);int x = sizeof(float); // block size = 4int convertedSize = wavSize / x;Int32 maxValue = Int32.MaxValue;float[] data = new float[convertedSize];int offset = 0;int i = 0;while (i < convertedSize){offset = i * x + headerOffset;data[i] = (float)BitConverter.ToInt32(source, offset) / maxValue;++i;}Debug.AssertFormat(data.Length == convertedSize, "AudioClip .wav data is wrong size: {0} == {1}", data.Length, convertedSize);return data;}#endregion

最终播放代码：

m_XunFeiTemplate.SetSpeaker(Speeker.XiaoYan_Woman_Mandarin);MemoryStream memoryStream = m_XunFeiTemplate.SpeechSynthesis(text.text);byte[] datas = memoryStream.ToArray();m_AudioSource.clip = ToAudio.Clip(datas);m_AudioSource.loop = false;m_AudioSource.volume = 1;m_AudioSource.Play();

有兴趣的小伙伴可以关注一波

o(*￣▽￣*)ブ