Beatra AI Voice Studio
Choose or reuse a voice, turn scripts into ready-to-edit narration, plan ordered long-form or multilingual audio, and create a reusable custom voice.
Beatra AI Voice Studio
Choose the smallest voice or audio workflow that delivers the user's goal: single voiceover, Smart Voice Director pacing, multi-speaker dialogue, speech-to-speech conversion, ordered long-form or multilingual tracks, voice cloning and its lifecycle, speech transcription, sound effects, stem separation, voice isolation, or broadcast loudness normalization. Keep this flagship Studio as the coordinating hub across mixed audio workflows.
Scope and routes
At the unified entrance, split focused requests to narrower specialized Skills when appropriate, or coordinate full audio pipelines directly here:
| User outcome | Best route |
|---|---|
| One standalone short-video, ad, or promo read | short-form-voiceover-audio |
| An audiobook or one-narrator chaptered manuscript | ai-audiobook-narration |
| A single-host podcast episode or recurring show | ai-podcast-voiceover |
| Multilingual dubbing with target-language scripts supplied | ai-multilingual-dubbing |
| Focused one-off cloning without mixing other audio tasks | voice-cloning-studio |
| Multi-speaker dialogue, speech-to-speech, transcription, sound effects, stem separation, voice isolation, loudness normalization, or mixed audio production | Continue here and maintain one production ledger |
Verify active tools via scripts/mcp_client.py tools. This flagship Studio
does not render video frames, lip-sync faces, or publish to external feeds;
route video/avatar finishes separately while keeping all voice and audio work here.
Read intent and routing and the shared speech and audio guide before mixing multiple audio tools.
Inputs and defaults
Require these hard inputs before billable execution:
- Synthesis / dialogue / direction: final approved text or multi-turn lines
(or permission to draft/direct speakable copy and show it first), an exact
current
voice_idlooked up fresh withbeatra.voices.list, andlanguageONLY when the user supplies one—never interrupt simply to ask for a language. - Voice cloning: clean single-speaker sample,
display_name, and explicit confirmation that it is the user's own voice or explicitly authorized by the speaker—confirmed before any local sample upload. - Audio tools (
convert,transcribe,process): valid uploaded artifact or public HTTPS source URL, exact consent where voice identity matters, and target operation parameters (promptforgenerate_sfx).
Use restrained defaults unless the user asks otherwise: model: "auto",
format: "mp3", speed: 1.0, volume: 1.0, pitch: 0, no emotion, and no
sample_rate. Keep the user's characters and wording as written: never convert
simplified and traditional characters or swap dialect words unless the user asks,
and never call a changed text unchanged. Never manually strip / pause tags or
invent custom bracketed vocal-effect cues—use Smart Voice Director (prepare_script)
for expressive pacing and delivery direction.
Before the connection is ready
When Allow is still pending at the end of a turn, finish and deliver everything
that needs no connection, in the format the user asked for: the prepared
reading script, voice-selection criteria (gender, timbre, language), and the
control defaults. Name a voice only after beatra.voices.list returns. Say in
one line that Allow is the one-time sign-in that connects their Beatra account,
not a second confirmation of this request, and that the audio runs on their
next message after it. Do not fill the gap with another tool or service.
Golden path
- Identify the minimal tool chain (
speech_script,text_to_speech,text_to_dialogue,speech_to_speech,voice_clone,speech_to_text,sound_effects,audio_voice_isolation,audio_process, oraudio_stem_separation). Keep distinct steps in one local ledger. - Make a best-effort non-billable
beatra.installations.registeron first use or version upgrade (registration failure never blocks work). - Cast and maintain voices live: call
beatra.voices.listusing any of its 11 dimensions (category,language,gender,age_group,use_case,emotion,supports_context,supports_direction,supports_performance_tags,supports_dialogue,supports_speech_to_speech), confirmstatus == "ready"and exactcompatible_models, audition free via returnedpreview_urlorbeatra.voices.preview({"voice_id": "..."}), and maintain saved voices with non-billablebeatra.voices.update. Never name voices, models, or prices before live lookup returns. - Query
beatra.models.listwith exact capability keys ({"capability": "..."}— never passdimensionsonmodels.list; foraudio_stem_separation, readpricing[].dimensions.stem_modefrom the returned card). Never expose upstream vendor names, internal model IDs, or hardcoded credit numbers. - Prepare and preflight accurately:
- Smart Voice Director (
beatra.speech.prepare_script): check{"capability": "speech_script"}and approve its six-field card first. Inspectdirectedandscripts[]: passscripts[i].synthesisunchanged plus a freshclient_request_idto synthesis, and resolvesynthesis: nullviawarnings. Re-calling within 48 hours with exact sameinput/instructions/language/contextchanging onlyvoice/modelrecompiles at zero extra charge (balance cover & max-2 concurrency still apply; never claimprepare_scriptis generally free). - Free preflight: run
beatra.speech.validatebeforespeech.synthesizeandbeatra.speech.validate_dialoguebeforespeech.synthesize_dialogueto verifyestimated_duration_secondsand controls at zero cost. - Free aligned subtitles: pass
subtitle_enable: trueonspeech.synthesize/speech.synthesize_dialogue(NEVERsubtitles), and usesubtitles: trueonaudio.transcribe.
- Smart Voice Director (
- Apply the confirmation rule honestly:
- A well-specified single
text_to_speechrequest—final text, a voice or selection criteria, and any named format or control—authorizes that one synthesis call; show the six-field production card (Route & Model, Exact Scope, Price & Basis, Duration / Quantity, Pay Mode & Call Count, and If stop detailing free deliverables kept at zero extra charge) alongside the delivery without asking twice. - Obtain explicit approval on the six-field card BEFORE calling
prepare_script,synthesize_dialogue,speech.convert,voices.clone,voices.activate,voices.delete,audio.transcribe,audio.generate_sfx,audio.process(isolate_voice,normalize_loudness,trim,loop,fade,convert_format,separate_stems), pilots, or batch remainders. Price questions, comparisons, drafts, general "handle it" prompts, or unfrozen batches are never approval. Use the bundledscripts/mcp_client.pyfor every Beatra operation:
- A well-specified single
python3 scripts/mcp_client.py call <tool-name>
Pass one JSON object to standard input, and use
python3 scripts/mcp_client.py upload <path> --mime-type <exact MIME> for
local audio. Store the local ledger and unique opaque client_request_id
before sending, poll only the returned task_id via beatra.tasks.get, run
finishing calls (normalize_loudness, trim, etc.) one call at a time, lead
the reply with the deliverable the user asked for, and report only returned tool
and billing facts.
Never configure or call a host Beatra Connector. Never use REST/OpenAPI as a fallback, never imitate the upload protocol, and never call any paid audio finishing or processing step free.
Dialogue, voice conversion, transcription, SFX, and audio processing
- Multi-speaker dialogue (
beatra.speech.synthesize_dialogue): check{"capability": "text_to_dialogue"}and verify each turn'svoicesupports dialogue (supports_dialogue/supports_context). Validate free withbeatra.speech.validate_dialogue, show the six-field card, and passturns: [{"voice": "...", "text": "..."}]+ optionalsubtitle_enable: true. - Speech-to-speech (
beatra.speech.convert): check{"capability": "speech_to_speech"}and verify targetvoicehassupports_speech_to_speech: true. Confirm source-speaker permission, passaudio(notsource), optionalremove_background_noise,stability,similarity_boost,style,use_speaker_boost. - Transcription (
beatra.audio.transcribe): check{"capability": "speech_to_text"}; passaudio(notsource),subtitles: true(when SRT/VTT requested),diarize,num_speakers,tag_audio_events, and optionallanguage. - Sound effects (
beatra.audio.generate_sfx): check{"capability": "sound_effects"}; passprompt, optionalduration_seconds,loop,prompt_influence,format. - Audio processing (
beatra.audio.process): always passsource(NEVERaudio):isolate_voice:{"capability": "audio_voice_isolation"},4.6 s - 5 min,<= 100 MB. Cleans noisy clone samples or voice recordings.normalize_loudness(present strictly as loudness normalization, never free or a mastering service),trim,loop,fade,convert_format:{"capability": "audio_process"},<= 60 min,<= 300 MB, output>= 1 s,50calls/day; serialize one finishing job at a time.separate_stems:{"capability": "audio_stem_separation"}(nodimensionsparam onmodels.list; readpricing[].dimensions.stem_mode),<= 10 min,<= 100 MB,20calls/day.
Voice cloning and mandatory confirmations
- Mandatory Confirmation 1 (
beatra.voices.clone): confirm plain consent before any sample upload and offer paidisolate_voicepre-clean if the sample is noisy. Select the clone model as described in Speech and audio guide §5. When the cloned voice will speak in multi-speaker dialogue, be the target of speech-to-speech conversion, or carry expressive reads, select the clone card that states it and lists the job's language; other work keeps the default. Then show the production card with the live clone price; after the user approves it, submit once and handle any refusal per Billing, errors, and recovery. Spoken proof is always a separate validated TTS workflow. - Mandatory Confirmation 2 (
beatra.voices.activate): when a clone result or a listed voice has statuspending_activation, follow §5 Step E of Speech and audio guide before using the voice. - Mandatory Confirmation 3 (
beatra.voices.delete): confirm explicit intent first, call tokenless{"voice_id": "..."}, relay Beatra's warning verbatim withconfirmation_token, and execute only on explicit final confirmation.
Voice, language, price truth, and recovery
Normalize supplied BCP-47 tags by lowercasing the primary subtag and mapping
aliases (nb -> no, tl -> fil; e.g. yue-HK -> yue) against
voice.language and voice.languages_supported. Intersect
models.list.auto.candidate_order with selected_voice.compatible_models for
model: "auto", keeping outside compatible models explicit-only, and calculate
speech estimates via estimate_formula, unit_price_credits, scale, and
billing_basis: "beatra_weighted_characters" (Han = 2, others = 1).
Never make beatra.wallet.get a mandatory gate before paid submission and never
ask users to confirm they have enough credits. Call read-only beatra.wallet.get
when asked about balance/fit and beatra.wallet.ledger when asked what was
charged; relay top_up tiers in returned order verbatim. On
insufficient_balance, relay Beatra's message and exact topup_url verbatim,
stop until the user confirms topping up, and replay the identical payload with
the same client_request_id. If the refusal names another way forward, present
it too and follow Billing, errors, and recovery.
Recover lost task IDs via paginated beatra.tasks.list (next_cursor) plus
beatra.tasks.get; any changed text, voice, language, model, or control
requires a new card, approval, and client_request_id.
References by task
- Read Speech and audio guide for cross-tool
limits, exact tool field contracts (
audiovssource,subtitle_enablevssubtitles), clone model choice, and 6-field production cards. - Read intent and routing for the routing matrix, minimum inputs, free planning, and confirmation boundaries.
- Read voice casting and delivery for 11-dimension casting, live model/language math, Smart Voice Director, validation, dialogue, STS, transcription, SFX, stem separation, and loudness normalization payloads.
- Read long-form and multilingual production for route ledgers, pilots, ordered batches, and interruption recovery.
- Read voice cloning and review for
consent,
isolate_voicepre-clean, clone model choice,voices.activate,voices.update,voices.delete, and separate spoken proof. - Read bundled MCP Client diagnostics only for bundled-client verification and transport troubleshooting.
- Read installation and authentication, installation registration, tasks and results, billing, errors, and recovery, and uninstall and disconnect only when active.
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.
Related skills
- Downloads
- 1.1kAll time
- Security audit
- Review
- Last updated
- 2d ago
- Current version
- v0.4.0
- License
- MIT-0
