change_voice
ChatGPTReplace the voice in an audio or video recording with a Voices AI voice and return a link to the new audio (MP3). Words, timing and emotion of the original are kept; video input returns audio only. Costs 1,000 credits per minute of audio, up to 60 minutes. Provide the recording in exactly one way: "file" (a file the user attached in ChatGPT), "media_url" (a direct link to an audio or video file, not a share page), or "upload_id" (from create_upload_link, when the user has a file but no link). Short recordings return the audio directly; longer ones return an id to check with get_generation.
create_song
ChatGPTGenerate an original song with vocals (or an instrumental). Costs 1000 credits and produces two versions. Either describe the song in "prompt", or pass full "lyrics" plus a "style". Songs take a few minutes; check progress with get_song.
create_upload_link
ChatGPTCreate a secure page where the user can upload an audio or video file (up to 500 MB) for change_voice, isolate_voice or transcribe. Use it when the user has a file they cannot attach or link. Give the user the link, wait until they say the upload is done, then call the tool with the upload_id. Links expire after 60 minutes.
generate_speech
ChatGPTTurn text into speech with a Voices AI voice and return a link to the MP3. Costs 1 credit per character (turbo: 0.5). At most 10,000 characters per call.
get_account
ChatGPTShow the Voices AI credit balance and whether the account has a subscription.
get_generation
ChatGPTCheck a speech generation, voice change or voice isolation by id (from generate_speech, change_voice, isolate_voice or list_history). Returns its status and, once complete, a link to the audio.
get_song
ChatGPTCheck a song from create_song or list_history. Returns its status and, once complete, the audio and cover image.
get_transcription
ChatGPTCheck a transcription by id (from transcribe or list_history) and, once complete, return its transcript. Long transcripts come in pages: pass next_offset from the previous call to continue.
isolate_voice
ChatGPTRemove background noise, music and room echo from an audio or video recording so only the voice remains, and return a link to the cleaned audio (MP3). Costs 1,000 credits per minute of audio; recordings must be between 5 seconds and 60 minutes. Provide the recording in exactly one way: "file" (a file the user attached in ChatGPT), "media_url" (a direct link to an audio or video file, not a share page), or "upload_id" (from create_upload_link, when the user has a file but no link). Short recordings return the audio directly; longer ones return an id to check with get_generation.
list_history
ChatGPTList the most recent speech generations and voice changes, songs, transcriptions or voice isolations in this Voices AI account, newest first.
search_voices
ChatGPTFind voices in the Voices AI voice library and the user's own cloned voices (category "custom", listed first). Returns voice ids for generate_speech and change_voice, the voice language and category, whether it works with the voice changer, and for library voices a portrait image and a short sample to preview.
transcribe
ChatGPTTranscribe the speech in an audio or video recording and return the transcript. Costs 100 credits per minute of audio. Formats: "md" (default, paragraphs with timestamps), "txt" (plain text), "srt" or "vtt" (subtitles). Provide the recording in exactly one way: "file" (a file the user attached in ChatGPT), "media_url" (a direct link to an audio or video file, not a share page), or "upload_id" (from create_upload_link, when the user has a file but no link). Short recordings return the transcript directly; longer ones return an id to check with get_transcription. Long transcripts come in pages: call get_transcription with next_offset for the rest.