Dialogue Audio

Install the belt CLI skill: npx skills add belt-sh/cli

Dialogue Audio

Create realistic multi-speaker dialogue with Dia TTS via inference.sh CLI.

Quick Start

Requires inference.sh CLI (belt). Install instructions

belt login

# Two-speaker conversation
belt app run falai/dia-tts --input '{
  "prompt": "[S1] Have you tried the new feature yet? [S2] Not yet, but I heard it saves a ton of time. [S1] It really does. I cut my workflow in half. [S2] Okay, I am definitely trying it today."
}'

Speaker Tags

Dia TTS uses [S1] and [S2] to distinguish two speakers.

Tag	Role	Voice
[S1]	Speaker 1	Automatically assigned voice A
[S2]	Speaker 2	Automatically assigned voice B

Rules:

• Always start each speaker turn with the tag
• Tags must be uppercase: [S1] not [s1]
• Maximum 2 speakers per generation
• Each speaker maintains consistent voice within a session

Emotion & Expression Control

Dia TTS interprets punctuation and non-speech cues for emotional delivery.

Punctuation Effects

Punctuation	Effect	Example
.	Neutral, declarative, medium pause	"This is important."
!	Emphasis, excitement, energy	"This is amazing!"
?	Rising intonation, questioning	"Are you sure about that?"
...	Hesitation, trailing off, long pause	"I thought it would work... but it didn't."
,	Short breath pause	"First, we analyze. Then, we act."
— or --	Interruption or pivot	"I was going to say — never mind."

Non-Speech Sounds

Dia TTS supports parenthetical sound descriptions:

(laughs)      — laughter
(sighs)       — exasperation or relief
(clears throat) — attention-getting pause
(whispers)    — softer delivery
(gasps)       — surprise

Examples with Emotion

# Excited conversation
belt app run falai/dia-tts --input '{
  "prompt": "[S1] Guess what happened today! [S2] What? Tell me! [S1] We hit ten thousand users! [S2] (gasps) No way! That is incredible! [S1] I know... I still cannot believe it."
}'

# Serious/thoughtful dialogue
belt app run falai/dia-tts --input '{
  "prompt": "[S1] We need to talk about the timeline. [S2] (sighs) I know. It is tight. [S1] Can we cut anything from the scope? [S2] Maybe... but it would mean dropping the analytics dashboard. [S1] That is a tough trade-off."
}'

# Teaching/explaining
belt app run falai/dia-tts --input '{
  "prompt": "[S1] So how does it actually work? [S2] Great question. Think of it like a pipeline. Data comes in on one end, gets processed in the middle, and comes out transformed on the other side. [S1] Like an assembly line? [S2] Exactly! Each step adds something."
}'

Pacing Control

Pause Hierarchy

Technique	Pause Length	Use For
Comma ,	~0.3 seconds	Between clauses, list items
Period .	~0.5 seconds	Between sentences
Ellipsis ...	~1.0 seconds	Dramatic pause, thinking, hesitation
New speaker tag	~0.3 seconds	Natural turn-taking gap

Speed Control

• Shorter sentences = faster perceived pace
• Longer sentences with commas = measured, thoughtful pace
• Questions followed by answers = engaging back-and-forth rhythm

# Fast-paced, energetic
belt app run falai/dia-tts --input '{
  "prompt": "[S1] Ready? [S2] Ready. [S1] Let us go! Three features. Five minutes. [S2] Hit it! [S1] Feature one: real-time sync."
}'

# Slow, contemplative
belt app run falai/dia-tts --input '{
  "prompt": "[S1] I have been thinking about this for a while... and I think we need to change direction. [S2] What do you mean? [S1] The market has shifted. What worked last year... is not working now."
}'

Conversation Structure Patterns

Interview Format

belt app run falai/dia-tts --input '{
  "prompt": "[S1] Welcome to the show. Today we have a special guest. Tell us about yourself. [S2] Thanks for having me! I am a product designer, and I have been building tools for creators for about ten years. [S1] What got you started in design? [S2] Honestly? I was terrible at coding but loved making things look good. (laughs) So design was the natural path."
}'

Tutorial / Explainer

belt app run falai/dia-tts --input '{
  "prompt": "[S1] Can you walk me through the setup process? [S2] Sure. Step one, install the CLI. It takes about thirty seconds. [S1] And then? [S2] Step two, run the login command. It will open your browser for authentication. [S1] That sounds simple. [S2] It is! Step three, you are ready to run your first app."
}'

Debate / Discussion

belt app run falai/dia-tts --input '{
  "prompt": "[S1] I think we should go with option A. It is faster to implement. [S2] But option B scales better long-term. [S1] Sure, but we need something shipping this quarter. [S2] Fair point... what if we do A now with a migration path to B? [S1] That could work. Let us prototype it."
}'

Post-Production Tips

Volume Normalization

Both speakers should be at consistent volume. If one is louder:

# Merge with balanced audio
belt app run infsh/video-audio-merger --input '{
  "video_file": "talking-head.mp4",
  "audio_file": "dialogue.mp3"
}'

Adding Background/Music

# Merge dialogue with background music (no inference.sh app mixes audio; use ffmpeg locally)
ffmpeg -i dialogue.mp3 -stream_loop -1 -i background-music.mp3 -filter_complex \
  "[1]volume=0.15[bg];[0][bg]amix=inputs=2:duration=first" mixed.mp3

Segmenting Long Conversations

For conversations longer than ~30 seconds, generate in segments:

# Segment 1: Introduction
belt app run falai/dia-tts --input '{
  "prompt": "[S1] Welcome back to another episode..."
}'

# Segment 2: Main content
belt app run falai/dia-tts --input '{
  "prompt": "[S1] So let us dive into today s topic..."
}'

# Segment 3: Wrap-up
belt app run falai/dia-tts --input '{
  "prompt": "[S1] Great conversation today..."
}'

# Merge all segments (no inference.sh app concatenates audio; use ffmpeg locally)
ffmpeg -i segment1.mp3 -i segment2.mp3 -i segment3.mp3 -filter_complex \
  "[0][1][2]concat=n=3:v=0:a=1" conversation.mp3

Script Writing Tips

Do	Don't
Write how people talk	Write how people write
Short sentences (< 15 words)	Long academic sentences
Contractions ("can't", "won't")	Formal ("cannot", "will not")
Natural fillers ("So,", "Well,")	Every sentence perfectly formed
Vary sentence length	All sentences same length
Include reactions ("Exactly!", "Hmm.")	One-sided monologues
Read it aloud before generating	Assume it sounds right

Common Mistakes

Mistake	Problem	Fix
Monologues longer than 3 sentences	Sounds like a lecture, not conversation	Break into exchanges
No emotional variation	Flat, robotic delivery	Use punctuation and non-speech cues
Missing speaker tags	Voices don't alternate	Start every turn with [S1] or [S2]
Formal written language	Sounds unnatural spoken	Use contractions, short sentences
No pauses between topics	Feels rushed	Use ... or scene breaks
All same energy level	Monotonous	Vary between high/low energy moments

Related Skills

# ElevenLabs dialogue (22+ voices, voice direction)
npx skills add inference-sh/skills@elevenlabs-dialogue

npx skills add inference-sh/skills@text-to-speech
npx skills add inference-sh/skills@ai-podcast-creation
npx skills add inference-sh/skills@ai-avatar-video

Browse all apps: belt app list