feat: custom system prompt (#21)

Author: liou666

components.d.ts

@@ -11,6 +11,7 @@ declare module '@vue/runtime-core' {
   export interface GlobalComponents {
     Avatar: typeof import('./src/components/Avatar.vue')['default']
     Button: typeof import('./src/components/Button.vue')['default']
+    ElCascader: typeof import('element-plus/es')['ElCascader']
     ElSwitch: typeof import('element-plus/es')['ElSwitch']
     ElTooltip: typeof import('element-plus/es')['ElTooltip']
     InputKit: typeof import('./src/components/InputKit.vue')['default']

src/hooks/useSpeechService.ts

@@ -83,7 +83,6 @@ export const useSpeechService = ({ langs = <const>['fr-FR', 'ja-JP', 'en-US', 'z
     mediaRecorder = new MediaRecorder(stream)
 
     mediaRecorder.ondataavailable = (e) => {
-      console.log(chunks, 'c')
       chunks.push(e.data)
     }
 
@@ -103,12 +102,10 @@ export const useSpeechService = ({ langs = <const>['fr-FR', 'ja-JP', 'en-US', 'z
 
     isRecognizReadying.value = true
 
-    recognizer.value.canceled = () => {
-      console.log('Recognize canceled')
-    }
     recognizer.value.recognized = (s, e) => {
       console.log('Recognize result: ', e.result.text)
       cb && cb(e.result.text)
+      // isRecognizing.value = false
     }
     recognizer.value.recognizing = (s, event) => {
       console.log('Recognize recognizing', event.result.text)
@@ -128,9 +125,9 @@ export const useSpeechService = ({ langs = <const>['fr-FR', 'ja-JP', 'en-US', 'z
       isRecognizing.value = false
     }
     recognizer.value.startContinuousRecognitionAsync(async () => {
-      await audioRecorder()
       isRecognizing.value = true
       isRecognizReadying.value = false
+      await audioRecorder()
       console.log('Recognize...')
     },
     (error) => {

src/pages/Home/components/Content.vue

@@ -114,9 +114,9 @@ async function onSubmit(fromRecognize = false) {
     ...currentChatMessages.value,
     { content: message.value, role: 'user', audioBlob: fromRecognize ? await blobToBase64(audioBlob.value) : '' },
   ])
-  const tempCurrentChatMessages = currentChatMessages.value.map(x => ({ content: x.content, role: x.role })) // 发送的请求中需去除audioBlob
+  const tempCurrentChatMessages = chatMessages.value.map(x => ({ content: x.content, role: x.role })) // 发送的请求中需去除audioBlob
   const systemMessage = currentChatMessages.value[0]
-  const relativeMessage = [...tempCurrentChatMessages, { content: message.value, role: 'user' }].slice(-(Number(chatRememberCount.value))) // 保留最近的几条消息
+  const relativeMessage = tempCurrentChatMessages.slice(-(Number(chatRememberCount.value))) // 保留最近的几条消息
   const prompts = [systemMessage, ...relativeMessage] as ChatMessage[]
 
   message.value = ''

src/pages/Home/components/NewChat.vue

@@ -15,33 +15,46 @@ const currentAvatarIndex = ref(Math.random() * avatarList.value.length | 0)
 const store = useConversationStore()
 const { ssmlToSpeak, isSynthesizing, isPlaying } = useSpeechService({ isFetchAllVoice: false })
 const allLanguages = computed(() => [...new Set(allVoices.map(v => v.locale))].filter(l => Object.keys(supportLanguageMap).includes(l)))
-const selectLanguage = ref('')
-const filterVoices = ref<VoiceInfo[]>([])
-const selectVoiceName = ref('')
 const desc = ref('')
 const name = ref('')
 const rate = ref('1.0')
 const previewText = ref('polyglot is awesome!')
-const filterStyles = ref<string[]>([])
-const selectStyle = ref('Neural')
+const presets = ref('Act as if you are meeting someone for the first time. How would you introduce yourself and start a conversation?')
 
-const canAdd = computed(() => !!(selectLanguage.value && selectVoiceName.value && desc.value && name.value))
-
-onBeforeMount(() => {
-  selectLanguage.value = allLanguages.value[0]
-  changeSelectLanguage(selectLanguage.value)
-})
+const voiceValue = ref<string[]>(['en-US', 'en-US-JennyNeural', 'chat'])
 
-watch(selectLanguage, changeSelectLanguage)
+const selectLanguage = computed(() => voiceValue.value[0])
+const selectVoiceName = computed(() => voiceValue.value[1])
+const selectStyle = computed(() => voiceValue.value[2])
+const canAdd = computed(() => !!(selectLanguage.value && selectVoiceName.value && desc.value && name.value))
 
-function changeSelectLanguage(newSelectLanguage: string) {
-  filterVoices.value = allVoices.filter(v => v.locale === newSelectLanguage)
-  selectVoiceName.value = filterVoices.value[0]?.shortName
+interface Option {
+  label: string
+  value: string
+  children?: Option[]
 }
 
-watch(selectVoiceName, (n) => {
-  filterStyles.value = filterVoices.value.filter(v => v.shortName === n)[0]?.styleList || []
-  selectStyle.value = filterStyles.value[0] || 'Neural'
+const options = ref<Option[] >([])
+
+onMounted(() => {
+  allLanguages.value.forEach((item) => {
+    const children: Option[] = []
+    allVoices.forEach((v) => {
+      if (v.locale === item) {
+        children.push({
+          value: v.shortName,
+          label: `${v.gender === 1 ? '🧒🏻' : '👦🏻'} ${v.localName}`,
+          children: v.styleList?.map(x => ({ label: voiceStyleMap[x], value: x })) || [],
+        })
+      }
+    })
+
+    options.value.push({
+      value: item,
+      label: supportLanguageMap[item],
+      children,
+    })
+  })
 })
 
 const randomAvatar = getAvatarUrl(avatarList.value[Math.random() * avatarList.value.length | 0]) // 随机默认选择一个头像
@@ -60,7 +73,7 @@ const addChat = (event: any) => {
     rate: +rate.value,
     isDefault: false,
     voiceStyle: selectStyle.value,
-  })
+  }, presets.value)
   store.changeCurrentKey(uid)
   emits('close')
 }
@@ -74,16 +87,8 @@ const previewSpeech = () => {
 }
 </script>
 
-<script>
-
-</script>
-
 <template>
-  <div flex="~ col gap-3" items-center>
-    <!-- <div text-lg font-bold>
-      自定义对话
-    </div> -->
-
+  <div class="wrapper" flex="~ col gap-3" items-center>
     <div flex>
       <Avatar v-model:image-url="imageUrl" />
     </div>
@@ -96,32 +101,17 @@ const previewSpeech = () => {
       <input v-model="desc" type="text">
     </div>
     <div flex>
-      <label for="">语言</label>
-      <select v-model="selectLanguage">
-        <option v-for="item in allLanguages" :key="item" :value="item">
-          {{ supportLanguageMap[item] }}
-        </option>
-      </select>
-    </div>
-    <div flex>
-      <label for="">音色</label>
-      <select v-model="selectVoiceName">
-        <option v-for="item in filterVoices" :key="item.shortName" :value="item.shortName">
-          {{ `${item.locale} / ${item.gender === 1 ? 'Female' : 'Male'} / ${item.localName}` }}
-        </option>
-      </select>
+      <label for="">语音</label>
+      <div w-55 flex>
+        <ElCascader v-model="voiceValue" style="width: 220px;" :options="options" />
+      </div>
     </div>
     <div flex>
-      <label for="">风格</label>
-      <select v-model="selectStyle">
-        <option value="Neural">
-          正常
-        </option>
-
-        <option v-for="item in filterStyles" :key="item" :value="item">
-          {{ voiceStyleMap[item] }}
-        </option>
-      </select>
+      <label for="">语速</label>
+      <div w-55 flex>
+        <input v-model="rate" flex-1 type="range" step="0.1" min="0.1" max="2.0">
+        <span w-4 ml-1>{{ Number(rate).toFixed(1) }}</span>
+      </div>
     </div>
     <div relative center-y>
       <div flex>
@@ -137,18 +127,12 @@ const previewSpeech = () => {
         </button>
       </div>
     </div>
+
+    <!-- todo -->
     <div flex>
-      <label for="">语速</label>
-      <div w-55 flex>
-        <input v-model="rate" flex-1 type="range" step="0.1" min="0.1" max="2.0">
-        <span w-4 ml-1>{{ Number(rate).toFixed(1) }}</span>
-      </div>
+      <label for="">场景预设</label>
+      <textarea v-model="presets" :rows="3" placeholder="system prompt" />
     </div>
-    <!-- todo -->
-    <!-- <div flex>
-        <label center-y justify-end mr-2 for="">预设</label>
-        <textarea id="message" resize-none w-50 block p-2 text-sm placeholder="Write your thoughts here..." />
-      </div> -->
     <div center-y text-sm text-gray-500>
       <i inline-block w-4 h-4 m-1 cursor-auto i-ic:baseline-lightbulb />
       点击头像可更换头像
@@ -162,12 +146,30 @@ const previewSpeech = () => {
 
 <style scoped>
   label{
-    @apply center-y justify-end mr-2 w-20
+    @apply center-y justify-center  w-20
   }
   input{
-    @apply w-50 p-2
+    @apply w-50 p-2 text-[#222]
   }
   select{
-    @apply w-55 select-settings
+    @apply w-55 select-settings text-[#222]
+  }
+  textarea{
+    @apply resize-none w-50 font-sans block p-2  text-[#222]
+  }
+  textArea::-webkit-scrollbar{
+    width: 0;
+    height: 0;
+  }
+  .wrapper :deep(.el-input__wrapper){
+    box-shadow: 0 0 0 1px #777 inset;
+    padding-top: 3px;
+    padding-bottom: 3px;
+  }
+  .wrapper :deep(.el-input__inner){
+    color: #222;
+  }
+  .wrapper :deep(.el-input__wrapper):hover {
+    box-shadow: 0 0 0 1px #777 inset;
   }
 </style>

src/pages/Home/components/Tool.vue

@@ -55,7 +55,7 @@ watch([azureKey, azureRegion, ttsPassword], () => {
       <span>Setting</span>
     </div>
   </div>
-  <Modal v-model:visible="addVisible" :z-index="2" class="dark:bg-[#111111] bg-white" center max-w-120 p6 @close="closeNewChat()">
+  <Modal v-model:visible="addVisible" h-80vh :z-index="2" class="dark:bg-[#111111] bg-white" center max-w-120 p6 @close="closeNewChat()">
     <NewChat :all-voices="tempAllVoices as any" @close="addVisible = false" />
   </Modal>
 

src/stores/index.ts

@@ -1,5 +1,5 @@
 import { v4 as uuid } from 'uuid'
-import { generatePrompt, getAvatarUrl } from '@/utils'
+import { basePrompt, generatePrompt, getAvatarUrl } from '@/utils'
 
 const defaultConversations = [{
   key: uuid(),
@@ -79,12 +79,12 @@ export const useConversationStore = defineStore('conversation', {
     cleanCurrentConversations() {
       this.chatMessages(this.currentKey)!.chatMessages.length = 1
     },
-    addConversation(conversation: Omit<Conversation, 'chatMessages'>) {
+    addConversation(conversation: Omit<Conversation, 'chatMessages'>, systemPrompt?: string) {
       this.conversations.push({
         ...conversation,
         chatMessages: [{
           role: 'system',
-          content: generatePrompt(conversation.language),
+          content: systemPrompt ? basePrompt(conversation.language, systemPrompt) : generatePrompt(conversation.language),
         }],
       })
     },

src/utils/openAi.ts

@@ -126,6 +126,9 @@ export const parseOpenAIStream = (rawResponse: Response) => {
 export const verifyOpenKey = (key?: string | null) => key && ([41, 51].includes(key.length))
 
 export const generatePrompt = (language: string) => {
-  return `I want you to act as an ${language} speaking partner and improver. No matter what language I speak to you, you need to reply me in ${language}. I hope you keep your responses clean and limit your responses to 100 characters. I hope you are humorous and interesting. I hope you will ask me a question from time to time in your reply. Now let\'s start practicing. Remember, I want you reply me in ${language} and you are humorous and interesting`
+  return `I want you to act as an ${language} speaking partner and improver. No matter what language I speak to you, you need to reply me in ${language}. I hope you keep your responses clean and limit your responses to 80 characters. I hope you are humorous and interesting. I hope you will ask me a question from time to time in your reply. Now let\'s start practicing. Remember, I want you reply me in ${language} and you are humorous and interesting`
 }
 
+export const basePrompt = (language: string, prompt?: string) => {
+  return `${prompt}.Remember, I want you reply me in ${language}, I hope limit your responses to 80 characters. Now let\'s start practicing. `
+}