215 lines
5.4 KiB
Go
215 lines
5.4 KiB
Go
package tts
|
||
|
||
import (
|
||
"bytes"
|
||
"context"
|
||
"encoding/base64"
|
||
"encoding/json"
|
||
"fmt"
|
||
"io"
|
||
"net/http"
|
||
"strings"
|
||
"time"
|
||
|
||
"github.com/hhs/camtalk/internal/trace"
|
||
"github.com/hhs/camtalk/internal/util"
|
||
"go.uber.org/zap"
|
||
)
|
||
|
||
// MiMoService 基于 Xiaomi MiMo TTS API 的语音合成实现。
|
||
// 接口兼容 OpenAI chat/completions 格式,通过 messages 传递待合成文本与风格指令。
|
||
type MiMoService struct {
|
||
apiKey string
|
||
model string
|
||
voice string
|
||
endpoint string
|
||
timeout time.Duration
|
||
logger *zap.SugaredLogger
|
||
client *http.Client
|
||
}
|
||
|
||
// NewMiMoService 创建 MiMo TTS 服务。
|
||
// model、voice、endpoint 由 config 层保证非空。
|
||
func NewMiMoService(apiKey, model, voice, endpoint string, timeoutSec, httpClientTimeoutSec int, logger *zap.SugaredLogger) *MiMoService {
|
||
timeout := time.Duration(timeoutSec) * time.Second
|
||
if timeout <= 0 {
|
||
timeout = 5 * time.Second
|
||
}
|
||
httpClientTimeout := time.Duration(httpClientTimeoutSec) * time.Second
|
||
if httpClientTimeout <= 0 {
|
||
httpClientTimeout = 30 * time.Second
|
||
}
|
||
return &MiMoService{
|
||
apiKey: apiKey,
|
||
model: model,
|
||
voice: voice,
|
||
endpoint: endpoint,
|
||
timeout: timeout,
|
||
logger: logger,
|
||
client: &http.Client{Timeout: httpClientTimeout},
|
||
}
|
||
}
|
||
|
||
// mimoTTSRequest MiMo TTS API 请求体。
|
||
type mimoTTSRequest struct {
|
||
Model string `json:"model"`
|
||
Messages []mimoTTSMessage `json:"messages"`
|
||
Audio mimoTTSAudio `json:"audio"`
|
||
Stream bool `json:"stream"`
|
||
}
|
||
|
||
// mimoTTSMessage MiMo TTS 消息。
|
||
type mimoTTSMessage struct {
|
||
Role string `json:"role"` // "user"(风格指令)| "assistant"(待合成文本)
|
||
Content string `json:"content"`
|
||
}
|
||
|
||
// mimoTTSAudio MiMo TTS 音频配置。
|
||
type mimoTTSAudio struct {
|
||
Format string `json:"format"` // "mp3" | "wav" | "pcm16"
|
||
Voice string `json:"voice"` // 预置音色 ID
|
||
}
|
||
|
||
// mimoTTSResponse MiMo TTS 非流式响应。
|
||
type mimoTTSResponse struct {
|
||
Choices []struct {
|
||
Message struct {
|
||
Audio struct {
|
||
Data string `json:"data"` // base64 编码的音频数据
|
||
} `json:"audio"`
|
||
} `json:"message"`
|
||
} `json:"choices"`
|
||
}
|
||
|
||
// mimoTTSStreamResponse MiMo TTS 流式响应。
|
||
type mimoTTSStreamResponse struct {
|
||
Choices []struct {
|
||
Delta struct {
|
||
Audio struct {
|
||
Data string `json:"data"` // base64 编码的音频数据片段
|
||
} `json:"audio"`
|
||
} `json:"delta"`
|
||
} `json:"choices"`
|
||
}
|
||
|
||
// SynthesizeStream 实现 tts.Service。从 textStream 读取句子,逐句调用 MiMo TTS API。
|
||
func (m *MiMoService) SynthesizeStream(ctx context.Context, textStream <-chan string, opts Options) (<-chan Chunk, error) {
|
||
voice := opts.Voice
|
||
if voice == "" {
|
||
voice = m.voice
|
||
}
|
||
|
||
ch := make(chan Chunk, 4)
|
||
go func() {
|
||
defer close(ch)
|
||
|
||
for text := range textStream {
|
||
if text == "" {
|
||
continue
|
||
}
|
||
|
||
audio, err := m.synthesize(ctx, text, voice)
|
||
if err != nil {
|
||
log := trace.FromContext(ctx)
|
||
log.Warnw("mimo tts: synthesize failed",
|
||
"error", err,
|
||
"text_len", len(text),
|
||
"text_preview", util.Truncate(text, 100))
|
||
// 静默跳过,不中断整个流
|
||
continue
|
||
}
|
||
|
||
select {
|
||
case ch <- Chunk{Audio: audio, IsLast: true, Final: false}:
|
||
case <-ctx.Done():
|
||
return
|
||
}
|
||
}
|
||
|
||
// textStream 关闭,发送 Final 标记
|
||
select {
|
||
case ch <- Chunk{Audio: nil, IsLast: false, Final: true}:
|
||
case <-ctx.Done():
|
||
}
|
||
}()
|
||
|
||
return ch, nil
|
||
}
|
||
|
||
// synthesize 调用 MiMo TTS API 合成单个句子。
|
||
// 使用非流式调用,返回完整音频数据(base64 解码后)。
|
||
func (m *MiMoService) synthesize(ctx context.Context, text, voice string) ([]byte, error) {
|
||
// 单句超时
|
||
ctx, cancel := context.WithTimeout(ctx, m.timeout)
|
||
defer cancel()
|
||
|
||
// 构建 MiMo TTS 请求:文本放在 assistant 消息中
|
||
reqBody := mimoTTSRequest{
|
||
Model: m.model,
|
||
Messages: []mimoTTSMessage{
|
||
{
|
||
Role: "assistant",
|
||
Content: text,
|
||
},
|
||
},
|
||
Audio: mimoTTSAudio{
|
||
Format: "mp3",
|
||
Voice: voice,
|
||
},
|
||
Stream: false,
|
||
}
|
||
|
||
payload, err := json.Marshal(reqBody)
|
||
if err != nil {
|
||
return nil, fmt.Errorf("mimo tts: marshal request: %w", err)
|
||
}
|
||
|
||
url := strings.TrimRight(m.endpoint, "/") + "/chat/completions"
|
||
|
||
req, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(payload))
|
||
if err != nil {
|
||
return nil, fmt.Errorf("mimo tts: create request: %w", err)
|
||
}
|
||
req.Header.Set("Content-Type", "application/json")
|
||
req.Header.Set("api-key", m.apiKey)
|
||
|
||
resp, err := m.client.Do(req)
|
||
if err != nil {
|
||
return nil, fmt.Errorf("mimo tts: send request: %w", err)
|
||
}
|
||
defer resp.Body.Close()
|
||
|
||
if resp.StatusCode != http.StatusOK {
|
||
errBody, _ := io.ReadAll(resp.Body)
|
||
return nil, fmt.Errorf("mimo tts: api error (status %d): %s", resp.StatusCode, string(errBody))
|
||
}
|
||
|
||
// 非流式响应:解析 JSON,提取 base64 音频数据
|
||
respBody, err := io.ReadAll(resp.Body)
|
||
if err != nil {
|
||
return nil, fmt.Errorf("mimo tts: read response: %w", err)
|
||
}
|
||
|
||
var ttsResp mimoTTSResponse
|
||
if err := json.Unmarshal(respBody, &ttsResp); err != nil {
|
||
return nil, fmt.Errorf("mimo tts: unmarshal response: %w", err)
|
||
}
|
||
|
||
if len(ttsResp.Choices) == 0 {
|
||
return nil, fmt.Errorf("mimo tts: empty choices in response")
|
||
}
|
||
|
||
audioData := ttsResp.Choices[0].Message.Audio.Data
|
||
if audioData == "" {
|
||
return nil, fmt.Errorf("mimo tts: empty audio data in response")
|
||
}
|
||
|
||
// base64 解码音频数据
|
||
audio, err := base64.StdEncoding.DecodeString(audioData)
|
||
if err != nil {
|
||
return nil, fmt.Errorf("mimo tts: decode audio base64: %w", err)
|
||
}
|
||
|
||
return audio, nil
|
||
}
|