Bugs

I am using instant mode in tts stream api, it was expected to give first audio chunk in 200ms, but it is taking about 2 secs, which is much higher...Please find the implementation below async def stream_tts(self, message_text): client = AsyncHumeClient(api_key=HUME_API_KEY) voice_id = "87c85b13-9dab-4b4d-bdd1-70182301d671" utterances = [ PostedUtterance(text = message_text, speed = 1, voice = PostedUtteranceVoiceWithId( id=voice_id, provider="CUSTOM_VOICE") ), ] logging.info (f"Sending TTS request with voice ID: {voice_id}") async for snippet in client.tts.synthesize_json_streaming( utterances=utterances, num_generations=1, split_utterances=False, instant_mode=True, format=FormatWav() ): audio_data = base64.b64decode( snippet.audio ) logging.info (f"Received raw wav snippet of size {len(audio_data)} bytes") Can you please help me improve the latency for this

If EVI is falsely interrupted (eg by some background noise rather than the user voice), leads to a blank user message, then EVI doesn't continue until prompted again by the user. This was fixed on Thu Sep 19 - EVI will now continue naturally after these interruptions.

Starting from April 18th, I am seeing that TTS is returning WAV files with duplicated utterances, generally doubling the size and length of the audio file.

i created a custom voice and applied it to all chapters (text to speach). although the voice appears selected, i experience a random change between the custom created voice and another standar voice i did not select or chose at any time. i've tried to change between the voices to try to get my custom voice to get assigned but was not able to do it. and all this changes ended up consuming my character count credit... so wasted credits...

EVI 2 sometimes does not speak the transcript correctly. Sometimes, ah or ums in the transcript turn into screeching noises. In longer responses especially, EVI tends to drop the last phrase or sentence, which is present in the transcript but not spoken.

Sometimes EVI will detect an interruption when there really isn't one (eg just background noise or the user backchanneling), and sometime it fails to recover from these false positive interruptions. We are working on improvements to reduce these issues.

I pasted my text into a new chapter of my project and the Vince Douglas voice was selected by default. I can pick my own voices from there, but it has defaulted to every paragraph and I don't need the extra selection there in my project. It increases the odds I make a mistake and generate the wrong voice, and the color is very close to my second voice I use so that makes it hard to see if the wrong voice is being chosen.

As My Voice plays back, the audio drops from sentence to sentence.