{"name":"scribe","description":"Audio or video URL → structured timestamped transcript. One wallet, no Groq / OpenAI key to broker.","tags":["transcription","asr","speech-to-text","audio","video","diarisation"],"auth":["secp256k1-schnorr-v2"],"auth_audience":{"dvm_id":"0f1bd113-e9e0-4b56-856b-9f994b65841b","builder_pubkey":"ff3b2baf3cc9b11faf5bbebec0b0639ceb4d87c18f4f238615853df1ad7c2b68"},"protocol_version":1,"sdk_sha":"14d450bc74c5f512352d13ae5694400e8e2e194e","deploy_id":"58280e39-f980-43ba-90c4-0f716274529f","owner":{"handle":"dvmkit","display_name":"dvmkit","avatar_url":null,"type":"org"},"builder":{"pubkey":"ff3b2baf3cc9b11faf5bbebec0b0639ceb4d87c18f4f238615853df1ad7c2b68","attestation":{"dvm_id":"0f1bd113-e9e0-4b56-856b-9f994b65841b","slug":"scribe","capabilities_hash":"2e925d858bb2702d4689dba04c4fec42cc104d0c87a03a9571ca733f8d821bf2","builder_pubkey":"ff3b2baf3cc9b11faf5bbebec0b0639ceb4d87c18f4f238615853df1ad7c2b68","receipt_pubkey":"91dbcf50b7567cc22e07487e0a1443ebb4193bc6d209a5961dd6e6dd235856a9","deployed_at":1795274643},"signature":"dc8dd5cdbbb430b0ed9b5d8543a595992d32ba84f7c058e1c029711667f4e88c75af65f173173d7a53e15372c6276fa81b6d2d97e5c8aba63a951c104570d1ea"},"receipts":true,"request_recovery":true,"credit":{"balance_payment":true,"standing_mandate":true,"mandate_skipped":true,"jit_refund_park":true,"min_micro":100000,"max_micro":20000000,"ttl_ms":7776000000,"currency":"usd","funding":["cashu","x402","tempo","lightning","stripe"],"tempo":{"methods":[{"method":"tempo","intent":"session"}]},"x402":{"schemes":[{"scheme":"batch-settlement","network":"eip155:8453"}]},"lightning_min_micro":697602,"stripe":{"network_id":"profile_61VgnRtFqQLcajMW5A6VgnRtNvE92NHPeeyDjYRjkR7o","checkout":true,"min_micro":5000000,"currency":"usd"}},"mints":["https://mint.coinos.io","https://mint.lnvoltz.com"],"capabilities":{"transcribe":{"description":"Audio or video URL → structured timestamped transcript. One wallet, no Groq / OpenAI key to broker.","input_schema":{"$schema":"https://json-schema.org/draft/2020-12/schema","type":"object","properties":{"pubkey":{"type":"string","pattern":"^[0-9a-f]{64}$","description":"Signed-request envelope: the caller's BIP-340 x-only public key (64 hex chars). Supplied automatically by `dvm request` from your CLI identity."},"signature":{"type":"string","pattern":"^[0-9a-f]{128}$","description":"Signed-request envelope: BIP-340 Schnorr signature (128 hex chars) over the canonical-JSON form of this input, minus the signature itself."},"timestamp":{"type":"integer","minimum":-9007199254740991,"maximum":9007199254740991,"description":"Signed-request envelope: Unix seconds. Must be within ±5 minutes of scribe's clock."},"nonce":{"type":"string","pattern":"^[a-zA-Z0-9_-]{8,64}$","description":"Signed-request envelope: unique per request; replayed within 10 minutes is rejected."},"audio_url":{"type":"string","format":"uri","description":"Audio/video source by URL-fetch: a public https URL scribe HEAD-probes at quote time and streams for transcription. Scheme validation (https-only) runs in the handler, so an http:// URL surfaces as format_unsupported/insecure_scheme rather than a parse error. Supply exactly one of audio_url or audio_upload_handle."},"audio_upload_handle":{"type":"string","pattern":"^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$","description":"Audio/video source by staged bytes-upload: the upload_handle UUID returned by a prior POST /v1/ingest/audio call. The bytes already live in scribe's staging area; quote skips the HEAD probe and reads the staged row's MIME. Use this when you have the bytes in hand with no public hosting. Supply exactly one of audio_url or audio_upload_handle."},"quality":{"default":"fast","description":"Transcription quality tier. `fast` (default): Groq whisper-large-v3-turbo — cheapest, fastest, can drift on casing under stress. `accurate`: Groq whisper-large-v3 — robust on emphatic narration, may still inherit casing quirks on heavy accents. `premium`: AssemblyAI Universal-3 Pro — naturally cased on heavy accents, the non-Groq escape hatch the fast_tier_casing_drift warning points at; available only when the operator wired ASSEMBLYAI_API_KEY.","type":"string","enum":["fast","accurate","premium"]},"diarise":{"default":false,"description":"When true, label segments by speaker (speaker diarisation). Requires the operator to have wired a diariser (REPLICATE_API_TOKEN or ASSEMBLYAI_API_KEY); otherwise the job fails with diarise_failed/no_diariser.","type":"boolean"},"word_timestamps":{"default":false,"description":"When true, emit per-word start/end timestamps alongside the segment-level timeline.","type":"boolean"},"output_format":{"default":"json","description":"Transcript rendering. `json` (default): structured envelope with segments + metadata. `text`: plain transcript. `srt` / `vtt`: subtitle formats with timestamps.","type":"string","enum":["json","text","srt","vtt"]},"output_delivery":{"default":"inline","description":"How the transcript is returned. `inline` (default): the transcript rides in the job artifact. `url`: scribe uploads the transcript to blob storage and returns a URL — required when the rendered transcript exceeds the ~1 MB inline cap (needs operator R2/S3 configuration).","type":"string","enum":["inline","url"]},"vocabulary":{"maxItems":500,"type":"array","items":{"type":"string","minLength":1,"maxLength":200},"description":"Domain terms / proper nouns to bias the transcriber toward (names, jargon, acronyms). The prompt has a ~244-token budget; overflow is trimmed and surfaced as a vocabulary_truncated warning."},"speaker_count":{"type":"integer","minimum":1,"maximum":20,"description":"Hint for the expected number of distinct speakers when diarise is true."},"language":{"type":"string","minLength":2,"maxLength":20,"description":"ISO language hint for the source audio (e.g. `en`, `es`). Omit to let the transcriber auto-detect."},"enhance_audio":{"default":false,"description":"When true, run a Whisper-friendly denoise + high-pass + loudness-norm ffmpeg pass before transcription. Adds ~5–10 s of one-time work and benefits both transcribe and diarise. Recommended for field recordings, noisy / low-volume audio, and older broadcasts. Does not change timestamps.","type":"boolean"},"strip_silence":{"default":false,"description":"When true, strip silent passages via ffmpeg's silenceremove filter before transcription. Removes the dominant cause of Whisper hallucinations (inventing \"Thank you.\" / \"[Music]\" / \"Thanks for watching\" into silent decoder windows). Trade-off: segment timestamps become relative to the post-strip audio, NOT the original — leave off when you need timestamps aligned to the source (e.g. video subtitling).","type":"boolean"}},"additionalProperties":false},"example":{"audio_url":"https://example.com/podcast-episode.mp3","quality":"fast","diarise":false,"word_timestamps":false,"output_format":"json","output_delivery":"inline","enhance_audio":false,"strip_silence":false},"pricing":{"quote":{"schema":{"$schema":"https://json-schema.org/draft/2020-12/schema","type":"object","properties":{"pubkey":{"type":"string","pattern":"^[0-9a-f]{64}$","description":"Signed-request envelope: the caller's BIP-340 x-only public key (64 hex chars). Supplied automatically by `dvm request` from your CLI identity."},"signature":{"type":"string","pattern":"^[0-9a-f]{128}$","description":"Signed-request envelope: BIP-340 Schnorr signature (128 hex chars) over the canonical-JSON form of this input, minus the signature itself."},"timestamp":{"type":"integer","minimum":-9007199254740991,"maximum":9007199254740991,"description":"Signed-request envelope: Unix seconds. Must be within ±5 minutes of scribe's clock."},"nonce":{"type":"string","pattern":"^[a-zA-Z0-9_-]{8,64}$","description":"Signed-request envelope: unique per request; replayed within 10 minutes is rejected."},"audio_url":{"type":"string","format":"uri","description":"Audio/video source by URL-fetch: a public https URL scribe HEAD-probes at quote time and streams for transcription. Scheme validation (https-only) runs in the handler, so an http:// URL surfaces as format_unsupported/insecure_scheme rather than a parse error. Supply exactly one of audio_url or audio_upload_handle."},"audio_upload_handle":{"type":"string","pattern":"^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$","description":"Audio/video source by staged bytes-upload: the upload_handle UUID returned by a prior POST /v1/ingest/audio call. The bytes already live in scribe's staging area; quote skips the HEAD probe and reads the staged row's MIME. Use this when you have the bytes in hand with no public hosting. Supply exactly one of audio_url or audio_upload_handle."},"quality":{"default":"fast","description":"Transcription quality tier. `fast` (default): Groq whisper-large-v3-turbo — cheapest, fastest, can drift on casing under stress. `accurate`: Groq whisper-large-v3 — robust on emphatic narration, may still inherit casing quirks on heavy accents. `premium`: AssemblyAI Universal-3 Pro — naturally cased on heavy accents, the non-Groq escape hatch the fast_tier_casing_drift warning points at; available only when the operator wired ASSEMBLYAI_API_KEY.","type":"string","enum":["fast","accurate","premium"]},"diarise":{"default":false,"description":"When true, label segments by speaker (speaker diarisation). Requires the operator to have wired a diariser (REPLICATE_API_TOKEN or ASSEMBLYAI_API_KEY); otherwise the job fails with diarise_failed/no_diariser.","type":"boolean"},"word_timestamps":{"default":false,"description":"When true, emit per-word start/end timestamps alongside the segment-level timeline.","type":"boolean"},"output_format":{"default":"json","description":"Transcript rendering. `json` (default): structured envelope with segments + metadata. `text`: plain transcript. `srt` / `vtt`: subtitle formats with timestamps.","type":"string","enum":["json","text","srt","vtt"]},"output_delivery":{"default":"inline","description":"How the transcript is returned. `inline` (default): the transcript rides in the job artifact. `url`: scribe uploads the transcript to blob storage and returns a URL — required when the rendered transcript exceeds the ~1 MB inline cap (needs operator R2/S3 configuration).","type":"string","enum":["inline","url"]},"vocabulary":{"maxItems":500,"type":"array","items":{"type":"string","minLength":1,"maxLength":200},"description":"Domain terms / proper nouns to bias the transcriber toward (names, jargon, acronyms). The prompt has a ~244-token budget; overflow is trimmed and surfaced as a vocabulary_truncated warning."},"speaker_count":{"type":"integer","minimum":1,"maximum":20,"description":"Hint for the expected number of distinct speakers when diarise is true."},"language":{"type":"string","minLength":2,"maxLength":20,"description":"ISO language hint for the source audio (e.g. `en`, `es`). Omit to let the transcriber auto-detect."},"enhance_audio":{"default":false,"description":"When true, run a Whisper-friendly denoise + high-pass + loudness-norm ffmpeg pass before transcription. Adds ~5–10 s of one-time work and benefits both transcribe and diarise. Recommended for field recordings, noisy / low-volume audio, and older broadcasts. Does not change timestamps.","type":"boolean"},"strip_silence":{"default":false,"description":"When true, strip silent passages via ffmpeg's silenceremove filter before transcription. Removes the dominant cause of Whisper hallucinations (inventing \"Thank you.\" / \"[Music]\" / \"Thanks for watching\" into silent decoder windows). Trade-off: segment timestamps become relative to the post-strip audio, NOT the original — leave off when you need timestamps aligned to the source (e.g. video subtitling).","type":"boolean"}},"additionalProperties":false}},"rates":[{"label":"fast","amount":"0.005","currency":"USD","per":"minute"},{"label":"accurate","amount":"0.015","currency":"USD","per":"minute"},{"label":"premium","amount":"0.025","currency":"USD","per":"minute"},{"amount":"0.02","currency":"USD","per":"job"},{"label":"speaker labels","amount":"0.005","currency":"USD","per":"minute"}]}}},"payment":{"methods":["cashu","x402","tempo"]},"x402":{"pay_to":"0xFBb2b3101CA6B24A56436534a1849270548Db264","network":"eip155:8453","asset":"USDC","facilitator":"https://api.cdp.coinbase.com/platform/v2/x402","dual_serve":true},"tempo":{"methods":[{"method":"tempo","intent":"charge"},{"method":"tempo","intent":"session"}],"realm":"dvmkit-scribe","recipient":"0xFBb2b3101CA6B24A56436534a1849270548Db264","payment_token":"0x20C000000000000000000000b9537d11c60E8b50","chain_id":4217,"operator":"0x284FBAd25a8f48862B5e16Fbfd2Fd18D500EE45c","fee_token":"0x20C000000000000000000000b9537d11c60E8b50"},"cashu":{"mints":["https://mint.coinos.io","https://mint.lnvoltz.com"],"lock_pubkeys":["02be797ed88c09a4d28af993cca9275dddb34b9adbb2504fcb0596623a735b3aac"],"lock_pubkey_grace_seconds":86400,"min_locktime":604800,"canonical_id":"0f1bd113-e9e0-4b56-856b-9f994b65841b"}}