{ "text": "Hey everyone.Imagine you're in a massive meeting with20 people.Everyone is talking at once and you needa perfect transcript that knows exactly\nwho said what the moment they say it.Usually, that's expensive and slow.But today, Meta just changed the game.\nMeta Superintelligence Labs hasofficially launched News VoiceTranscribe.This is a brand new audio perception\nmodel that handles streamingtranscription and speaker diarizationfor over 20 people simultaneously.The most shocking part?\nThe price.Meta is offering this for just 18 centsper hour of processed audio.To put that in perspective, competitors\nlike AssemblyAI or Deepgram can cost twoto three times more when you add inspeaker identification.Muse isn't just cheap, it's smart.\nIt processes audio in tiny80-millisecond chunks using a trickcalled adaptive delay.It can wait a split second if the speech\nis messy or commit immediately if it'sclear.It supports over 70 languages andhandles things like code-switching where\npeople jump between languages in onesentence.Now, is it a world record?\nNot quite.Speechmatics can handle up to 100speakers and Amazon Transcribe handles30.But for most businesses building meeting\nassistants or call analytics, 20speakers is plenty.Plus, Meta is topping the charts foraccuracy.In recent benchmarks from Artificial\nAnalysis, Muse hit a 3.1% word errorate beating out rivals like OpenAI,Google, and ElevenLabs.It's also leading in diarization\naccuracy, which means fewer mistakes infiguring out who is talking.While there are some tradeoffs like no\nword-level timestamps yet, the valuehere is incredible.For enterprise developers, this is amassive win for building live AI\nassistants and meeting tools withoutbreaking the bank.Meta is making high-end voice AIaccessible to everyone.\nWhat do you think?\nIs 18 cents an hour low enough to changehow your company uses AI?\nLet me know in the comments.Like, subscribe, and turn onotifications.", "language": "en", "duration": 142.0, "segments": [ { "id": 0, "start": 0.0, "end": 9.84, "text": "Hey everyone.Imagine you're in a massive meeting with20 people.Everyone is talking at once and you needa perfect transcript that knows exactly", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 1, "start": 9.84, "end": 19.08, "text": "who said what the moment they say it.Usually, that's expensive and slow.But today, Meta just changed the game.", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 2, "start": 19.08, "end": 26.16, "text": "Meta Superintelligence Labs hasofficially launched News VoiceTranscribe.This is a brand new audio perception", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 3, "start": 26.16, "end": 34.88, "text": "model that handles streamingtranscription and speaker diarizationfor over 20 people simultaneously.The most shocking part?", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 4, "start": 34.88, "end": 43.84, "text": "The price.Meta is offering this for just 18 centsper hour of processed audio.To put that in perspective, competitors", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 5, "start": 43.84, "end": 53.6, "text": "like AssemblyAI or Deepgram can cost twoto three times more when you add inspeaker identification.Muse isn't just cheap, it's smart.", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 6, "start": 53.6, "end": 61.96, "text": "It processes audio in tiny80-millisecond chunks using a trickcalled adaptive delay.It can wait a split second if the speech", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 7, "start": 61.96, "end": 69.36, "text": "is messy or commit immediately if it'sclear.It supports over 70 languages andhandles things like code-switching where", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 8, "start": 69.36, "end": 75.4, "text": "people jump between languages in onesentence.Now, is it a world record?", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 9, "start": 75.4, "end": 85.32, "text": "Not quite.Speechmatics can handle up to 100speakers and Amazon Transcribe handles30.But for most businesses building meeting", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 10, "start": 85.32, "end": 95.56, "text": "assistants or call analytics, 20speakers is plenty.Plus, Meta is topping the charts foraccuracy.In recent benchmarks from Artificial", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 11, "start": 95.56, "end": 105.8, "text": "Analysis, Muse hit a 3.1% word errorate beating out rivals like OpenAI,Google, and ElevenLabs.It's also leading in diarization", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 12, "start": 105.8, "end": 112.64, "text": "accuracy, which means fewer mistakes infiguring out who is talking.While there are some tradeoffs like no", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 13, "start": 112.64, "end": 120.96, "text": "word-level timestamps yet, the valuehere is incredible.For enterprise developers, this is amassive win for building live AI", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 14, "start": 120.96, "end": 129.16, "text": "assistants and meeting tools withoutbreaking the bank.Meta is making high-end voice AIaccessible to everyone.", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 15, "start": 129.16, "end": 130.96, "text": "What do you think?", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 16, "start": 130.96, "end": 136.04, "text": "Is 18 cents an hour low enough to changehow your company uses AI?", "source": "video_source_subtitle", "authority": "video_source_subtitle" }, { "id": 17, "start": 136.04, "end": 142.0, "text": "Let me know in the comments.Like, subscribe, and turn onotifications.", "source": "video_source_subtitle", "authority": "video_source_subtitle" } ], "transcript_selection": { "schema": "videonote.transcript_selection.v1", "status": "SELECTED", "requested_language": "en", "duration": 142.0, "reference_path": "/Users/sagawa/AI_WORK/video_notes/url/20260914_194308__Meta_Just_Broke_the_Cost_of_AI_Transcription_Muse_Voice_Transcribe_Is_Here/reference_subtitle.vtt", "reference_kind": "automatic_original", "reference_language": "en-orig", "selected_source": "video_source_subtitle", "source_caption_kind": "automatic_original", "source_caption_processing": "rolling_dedup_and_sentence_merge", "youtube_overlap_clamps": [], "general_high_confidence_repairs": [], "decision_reason": "complete_source_subtitle_fast_path_no_whisper", "whisper_rejected": false, "source_profile": { "segment_count": 18, "duration_coverage": 1.0 } } }