Interactive Avatar streaming , voice chat

Hi, i have a problems during attempt to integrate interactive avatar.

There are my expectations about how that process should work, very simple version

Initialize avatar
Ask GPT about opening phrase (by me, i have a backend that communicates with GPT)
Avatar to say GPT response
Listening to user input, transcribe it (both by heyGen sdk, im only getting transcription from sdk)
Ask GPT about response based on what user said (again me and my backend)
Go 3 and repeat infinitely.

Problem that i have related to step 4. The problem with my expectation OR with a code i did, so i need someone's clarifications please.

Code below omits some steps of list above, this code should just start avatar and listen to user voice (expected to).

  async startSession(videoElement: HTMLVideoElement): Promise<void> {
    const token = await this.fetchAccessToken();

    this._Avatar = new StreamingAvatar({ token });

    this._Avatar.on(StreamingEvents.STREAM_READY, (x) => {
      videoElement.autoplay = true;
      videoElement.playsInline = true;
      videoElement.muted = false;
      videoElement.srcObject = this._Avatar.mediaStream;

    });
 
    const sessionData = await this._Avatar.createStartAvatar({
      avatarName: 'Wayne_20240711',
      quality: AvatarQuality.High,
      language: "English",
    });

    await this._Avatar.startVoiceChat({ useSilencePrompt: false });
    //await this._Avatar.startListening() 

  }

when execution reaches that line

await this._Avatar.startVoiceChat({ useSilencePrompt: false });

error appears in console, it related to fail of WS connection start

wss://api.heygen.com/v1/ws/streaming.chat?session_id=9fb858d0-acc9-11ef-92b7-ee31df050ca4&session_token=eyJ0b2tlbiI6ICIzN2Y5MTAyMWRlYzU0Nzc1YTk4NmE2Mjk3YTNjNjY0YiIsICJ0b2tlbl90eXBlIjogInNhX2Zyb21fcmVndWxhciIsICJjcmVhdGVkX2F0IjogMTczMjcxNjc0M30=&silence_response=false&stt_language=English

I expect that after fixing that issue (when WS connection start wont fail) - i will be able to recieve transcription via appropriate event of avatar object, like that

this.Avatar.on(StreamingEvents.USER_END_MESSAGE,(event:UserTalkingEndEvent)=>{
  //event?.details
  //event?.text
  //event?.transcription
  //event?.whatever
})

Tell me please if im wrong and where, and if so - how I would achieve desired behavior? I know about "knowledge id\base" approach with custom avatar, but as i understand from customer support explanation my approach and "knowledge" is 2 different but possible ways of making it work