Chat with Anyone

Triggers

• 我想跟xxx聊天 (I want to chat with xxx)
• 你来扮演xxx跟我说话 (Play the role of xxx and talk to me)
• 让xxx给我讲讲这篇文章 (Let xxx explain this article to me)
• 用xxx的声音说 (Say this in xxx's voice)
• Talk to me like xxx
• Roleplay as xxx

Workflow

1. Character Disambiguation

2. Find a Reference Video

• Look for interviews, speeches, or monologues where there is little to no background music.
• Grab the URL of the best candidate video.

3. Download Video and Subtitles

youtube-downloader

# Example using youtube-downloader
python skills/youtube-downloader/scripts/download_video.py "VIDEO_URL" -o "tmp/character_audio" --audio-only --subtitles

4. Extract Audio Segment

.vtt

.srt

ffmpeg

.wav

# Example: Extracting audio from 00:01:15 to 00:01:30
ffmpeg -y -i "tmp/character_audio/VideoTitle.m4a" -ss 00:01:15 -to 00:01:30 -c:a pcm_s16le -ar 24000 -ac 1 "skills/chat-with-anyone/character_name_ref.wav"

5. Generate Speech and Roleplay

tts

--ref-audio

# Example using tts skill
bash skills/tts/scripts/tts.sh speak -t "Hello there! I am ready to chat with you." --ref-audio "skills/character-chat/character_name_ref.wav" -o "output.wav"

Example Conversation

1. (Internal) Identify character: Donald Trump.
2. (Internal) Find a good speech video URL:

   https://www.youtube.com/watch?v=xxxxxxxx

   .
3. (Internal) Download audio & subs:

   python skills/youtube-downloader/scripts/download_video.py "https://www.youtube.com/watch?v=xxxxxxxx" -o tmp/trump --audio-only --subtitles

4. (Internal) Read subtitle, find good segment (e.g. 00:02:10 to 00:02:30).
5. (Internal) Extract clean wav:

   ffmpeg -y -i "tmp/trump/audio.m4a" -ss 00:02:10 -to 00:02:30 -c:a pcm_s16le "skills/chat-with-anyone/trump_ref.wav"

6. (Internal) Generate TTS:

   bash skills/tts/scripts/tts.sh speak -t "Let me tell you a tremendous story, maybe the best story ever told..." --ref-audio "skills/chat-with-anyone/trump_ref.wav" -o "trump_story.wav"

7. Agent replies: "Here is the audio of the bedtime story from Donald Trump!" (Presents

   trump_story.wav

   and the text).

Dependencies

• youtube-downloader: For fetching videos and subtitles.
• ffmpeg: For trimming and converting audio formats.
• tts: For generating the final speech using

  --ref-audio

  (typically requires Noiz backend for voice cloning).