Skip to content

beatra-ai/beatra-skills / skills/audio-drama-voice-studio

audio-drama-voice-studio

Verified

Give every character in your script its own voice and get the whole scene recorded line by line, in script order, as numbered clips ready to drop into your editor. This AI multi-character voice generator covers audio drama and radio play, dramatised fiction and audiobooks with a cast, dialogue voice over, two-host and panel shows, scripted ads with two roles, and course roleplay scenes: decide who is who first — age, temperament, how far apart the voices should sit — give the narrator a voice of their own, and afterwards re-record just the one line you want to change. Also used for audio drama voice generator work, multi-character voice generator work, multi-speaker audio, dialogue text to speech, script to dialogue audio, AI audio drama, radio play voices, and full cast voiceover.

Audio Drama Voice Studio

Give a script a cast. Assign one voice per character, plan expressive pacing without rewriting dialogue, record scenes natively with multi-speaker dialogue or block-by-block synthesis, add scene foley/ambience and -14 LUFS finishing when requested, and hand back ordered clips and aligned subtitles ready to assemble in your editor.

What you get

A finished run returns:

  • Numbered scene takes (beatra.speech.synthesize_dialogue) or per-block clips (beatra.speech.synthesize), labelled with speaking roles in script order
  • A cast sheet: each speaker, age/temperament, how they contrast relative to the others, and the opaque voice_id kept for the whole piece
  • Free aligned subtitles (output.subtitles: JSON, SRT, VTT, and dialogue segments[]) when subtitle_enable: true (default; never pass subtitles)
  • Optional separately confirmed foley/ambience beds (beatra.audio.generate_sfx) and finished clips (beatra.audio.process)
  • Actual returned duration, size, MIME type, resolved model, and billing.net_charged_credits for every delivered clip

It does not mix multiple audio tracks into one layered episode master. Multi-track timeline mixing and SFX layering stay in the user's editor.

Ways to run it

  • Cast, direct, and record — default. Build the cast sheet, optionally run Smart Voice Director (beatra.speech.prepare_script), audition free beatra.voices.preview clips, validate free with beatra.speech.validate_dialogue or beatra.speech.validate, then synthesize scenes via beatra.speech.synthesize_dialogue (or segmented beatra.speech.synthesize when cast limits or isolated-line control require it).
  • Add foley, ambience, and loudness finishing — optional, separately quoted. Generate scene foley, stingers, or loop=true ambient beds (beatra.audio.generate_sfx) and run beatra.audio.process (isolate_voice, normalize_loudness, trim, fade). Never call any paid finishing or script-direction step free.
  • Add an owned voice — optional. Pre-clean noisy samples if needed (isolate_voice), then clone a voice with explicit owner consent and full lifecycle confirmations (consent_attested, beatra.voices.activate when pending_activation, two-step beatra.voices.delete).

Casting, script drafting, beatra.voices.list, beatra.voices.preview, beatra.models.list, beatra.speech.validate, beatra.speech.validate_dialogue, and beatra.voices.delete step 1 are free. Smart Voice Director (beatra.speech.prepare_script), synthesis, dialogue synthesis, cloning, activation, SFX, and beatra.audio.process are paid steps.

Scope and routing

Use this Skill when a script or dialogue passage needs more than one speaker: audio drama and radio play, dramatised fiction, audio novels with named roles, a two-host or panel show, a dialogue-driven course or training scene, a scripted ad with two or more roles, or an interactive story with distinct characters.

Route a single-narrator book or course to ai-audiobook-narration. Route a podcast episode to ai-podcast-voiceover, and a short-video voiceover to short-form-voiceover-audio. Route standalone cloning-only work to voice-cloning-studio when cloning is the whole job; if a character on this cast needs a cloned voice, clone and manage it right here. Route general single-voice work to beatra-ai-voice-studio, and narrated video with pictures to novel-promo-video-maker.

Inputs and defaults

The one hard input is the script text. Reuse the genre, era, tone, character notes, language, and any voice already chosen in the conversation.

Ask only when the answer changes the paid result: who speaks each line, when the passage is prose with no speaker attribution and the text does not settle it.

Defaults that avoid extra questions:

  • One opaque voice_id per named character, plus one narrator voice when the script has narration.
  • Speaker order follows the script; nothing is reordered.
  • model omitted or "auto" unless a capability requires an explicit choice.
  • language omitted for automatic detection unless the user specifies one or the script mixes several.
  • subtitle_enable: true (never pass subtitles) to obtain free timestamps and aligned JSON/SRT/VTT subtitles (segments[] on dialogue).
  • Use native beatra.speech.synthesize_dialogue when the scene fits 2–50 turns, <= 2,000 characters, <= 10 voices with capabilities.supports_dialogue == true, and <= 4 ready clones (chunk longer scenes cleanly with previous_text / next_text). Fall back to segmented beatra.speech.synthesize when cast limits, dialect constraints, or single-line isolation apply.

Golden path

  1. Read the script and build the cast sheet: every speaking character, age, temperament, relative contrast, and narrator if present. Split the script into speaker-attributed turns/scenes (noting <= 2,000-char dialogue scenes for native beatra.speech.synthesize_dialogue or individual blocks for beatra.speech.synthesize). Show the numbered cast sheet and script layout to the user as the free visible deliverable. Planning is not approval.
  2. Call beatra.voices.list (bilingual search, limit <= 500, cursor, facet filters, supports_dialogue=true when targeting dialogue, inspecting all 11 capabilities.* metadata fields) only when voices are still needed or the user asks to browse/compare. Run free beatra.voices.preview (preview_url, kind) to audition picks and freeze one opaque voice_id per role.
  3. Optional Smart Voice Director: if dramatic beats, pause holds, or expressive timing need compilation without changing dialogue words (never hand-author braced tags), quote and run paid beatra.speech.prepare_script ({"capability": "speech_script"}, max 2 concurrent; 48-hour same-script voice swap reuse when balance covers admission), inspect directed and scripts[] (resolving any synthesis: null from warnings), then pass ready scripts[].synthesis unchanged + fresh client_request_id.
  4. Read live cards via beatra.models.list (text_to_dialogue, text_to_speech, speech_script, sound_effects, audio_process, audio_voice_isolation) and run free preflight (beatra.speech.validate_dialogue for dialogue scenes or beatra.speech.validate for single blocks) for live pricing, duration estimates, weighted characters (Han ideographs = 2, others = 1; linear per 1,000 with exact fractional charge under 1,000), and warnings.
  5. Show one current six-field production card covering the approved scene/blocks (plus separate cards or itemized paid lines for any optional SFX or finishing steps) and wait. Never treat free planning as approval.
  6. Submit each approved call once via bundled scripts/mcp_client.py with its own opaque client_request_id (beatra.speech.synthesize_dialogue or beatra.speech.synthesize, plus optional beatra.audio.generate_sfx or beatra.audio.process) and poll beatra.tasks.get until terminal.
  7. Deliver ordered clips, speaker labels, free aligned subtitles (segments[]), actual duration, size, MIME type, artifact ID/URL, resolved model, and billing.net_charged_credits. Review only audio the host can actually play.

A single-line revision re-runs only the affected line or scene chunk under a new production card and client_request_id, leaving accepted clips untouched. See the audio drama workflow and speech, voices, cloning, and audio finishing.

Decisions that require confirmation

Planning, cast sheets, voice previews, and preflight quotes are free and never constitute approval. Before the first paid call, show a six-field production card:

  1. Work — exact scene/block scope, voices, language/format/controls, and any requested prepare_script, audio.generate_sfx (prompt, duration_seconds 0.5–30.0, loop=true), or beatra.audio.process step (isolate_voice, normalize_loudness -14 LUFS, trim, fade).
  2. Credits — live rate/quote from beatra.models.list, beatra.speech.validate_dialogue, or beatra.speech.validate. Never guess or promise an estimate is final.
  3. Count — exact number of paid calls approval creates. A changed line or parameter is a new call.
  4. Identity — one fresh caller-generated opaque client_request_id per paid call. Never reuse an ID across clips or changed parameters.
  5. If we stop here — the cast sheet, script layout, audition picks, and any already-delivered clips remain usable at zero extra charge (or nothing paid yet before the first submit).
  6. If the balance is insufficient — relay Beatra's official message and exact top-up URL verbatim without modifying the URL, and wait until the user says they have topped up before retrying with the same frozen client_request_id.

Submit once through bundled scripts/mcp_client.py. Poll beatra.tasks.get.

Voice cloning requires three separate mandatory confirmations alongside its own card (voice_clone, voices.activate, voices.delete); a synthesis card never authorizes cloning, SFX, or finishing steps not listed on that card.

Using a cloned voice (3 mandatory confirmations)

Follow the full consent and activation flow in speech, voices, cloning, and audio finishing:

  1. Confirmation #1 — Consent (consent_attested: true): obtain explicit confirmation in this conversation that the user owns the voice or holds the speaker's permission before uploading via scripts/mcp_client.py upload. Possession of an audio file is never consent.
  2. Noisy sample check (paid): if a recording has music, street noise, or room echo, propose paid beatra.audio.process (operation="isolate_voice", {"capability": "audio_voice_isolation"}, 4.6 s – 5 min, <= 100 MB) on its own card first.
  3. Clone model choice: select the clone model as described in speech, voices, cloning, and audio finishing §5. When a role speaks in a multi-character dialogue scene or needs expressive delivery, select the voice_clone model card that states it; otherwise keep the default. Then present the six-field clone card with the live clone price; after the user approves it, submit beatra.voices.clone once and handle any refusal per billing, errors, and recovery.
  4. Confirmation #2 — Activation (beatra.voices.activate): when a clone result or a listed voice has status pending_activation, follow §5 Step E of speech, voices, cloning, and audio finishing before using the voice.
  5. Confirmation #3 — Two-step permanent deletion (beatra.voices.delete): when deleting a clone, first call beatra.voices.delete with voice_id only, display paid_slot_warning, irreversible_notice, and 15-minute token_expires_at, wait for explicit confirmation, then call step 2 with voice_id and confirmation_token.

Offer the free clone preview only when the user asks to hear the voice before paying, or when a paid clone is refused and the refusal offers it. Detailed payloads and sample bounds live in the audio drama workflow.

Account balance

When the user asks how many credits remain or whether a live estimate fits, call read-only beatra.wallet.get. When they ask what was charged, call beatra.wallet.ledger. Never make beatra.wallet.get a mandatory gate before submitting, and never ask the user to confirm they have enough credits. If a model card carries a top_up block, relay its tiers verbatim in listed order without ranking or picking one.

How this Skill executes

Use the bundled scripts/mcp_client.py for every remote Beatra operation: the MCP tool name is the CLI argument after call, and one JSON object goes on standard input. Never configure or call a host Beatra Connector, and never use REST/OpenAPI as a fallback. Register the package with beatra.installations.register on first use. Submit paid operations once, then poll to terminal state.

Recovery

Save every returned task_id immediately and poll with beatra.tasks.get; queued and running mean wait. Replay a create only when its response is genuinely unknown and the payload is identical field-for-field under the same client_request_id. If a task_id is lost, call beatra.tasks.list scoped to the exact capability (text_to_dialogue, text_to_speech, speech_script, voice_clone, sound_effects, audio_process, audio_voice_isolation), follow every next_cursor across the submission window, confirm candidates with beatra.tasks.get, and stop if ambiguous or if client_request_id itself was lost. Call beatra.tasks.cancel only on user request; on 409, keep polling the original task. Recover only failed or missing segments while keeping valid delivered clips untouched. See speech, voices, cloning, and audio finishing for exact structured error handling (voice_model_incompatible, daily_limit, insufficient_balance).

References by task

Updates and security

This Skill runs only the files installed with it and leaves them unchanged while it runs; it never downloads or executes other code. It talks to Beatra's official services over HTTPS, and a file you choose to upload goes only to the one-time upload address Beatra issues for it.

Once a day at most, the bundled client reads Beatra's public version list. When the store or tool you installed from has a newer version, it prints one line with the command that updates it there. Updating stays your choice, through that store or tool. See updates and security.

slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 196kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 196kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 196kUpdated Claude CodeCodexVerifiedInstall