NGSpeak: Language Learning App
Learn English and Bengali with 160 offline-first lessons across CEFR A1–C2, including speaking, writing, pronunciation practice, and progress tracking.
Download NowConvert existing NIST RTTM speaker diarization speech diarization annotations to YAML records locally without audio uploads.
Learn English and Bengali with 160 offline-first lessons across CEFR A1–C2, including speaking, writing, pronunciation practice, and progress tracking.
Download NowUp to 2 MiB · 1–1,000 turns, maximum 32 speakers · Local conversion
STEP BY STEP
FORMAT GUIDE
Review per-speaker speaking intervals in a spreadsheet or exchange simple RTTM and TextGrid annotations.
UNDERSTAND YOUR FILES
NIST RTTM uses ten-field SPEAKER turns for diarization. This tool accepts a restricted single-recording, channel-1, millisecond-precision profile.
YAML represents structured information in readable text. These tools support flat lists of named records rather than nested configuration files.
Review per-speaker speaking intervals in a spreadsheet or exchange simple RTTM and TextGrid annotations. Choosing an output format changes how another app reads the file; it does not restore detail missing from your source.
Check column headings, escaped characters, delimiters and empty values. Data type information and spreadsheet formulas may not survive a plain-text format change.
Your selected file is converted on your device. You can download the new file while keeping the original unchanged.
Some formats work differently across browsers and devices. Check the converted result before using it for important work. Advertising and other optional website services may have separate privacy practices.
Only a restricted NIST SPEAKER RTTM profile and a dedicated Praat TextGrid long-text profile are accepted. Required table columns in order: recording_id, speaker_id, start_ms, duration_ms. Supports one recording, 1–1,000 speech intervals, up to 32 speakers, a 24-hour maximum timeline, and exact millisecond timestamps. RTTM requires channel 1 and all four optional fields <NA>. TextGrid uses one IntervalTier per speaker, named RTTM:recording_id:speaker_id, with text "speech" during active intervals and empty text for gaps. Speakers may overlap across tiers, but not with themselves. Other Praat tiers, diarization metadata, transcripts, time values with more than three decimals, and nonconforming TextGrid formats are rejected. Conversions do not analyze speech or edit audio. Keep the originals and check exported annotations in the destination program.
Actual output behavior depends on the input file, your browser, device resources and the converter engine. Check important files before sharing or replacing them.
HELP & ANSWERS
No. It converts existing annotated intervals only.
recording_id, speaker_id, start_ms, duration_ms, in that order.
No. Only the documented RTTM speaker-tier profile is supported. Only a restricted NIST SPEAKER RTTM profile and a dedicated Praat TextGrid long-text profile are accepted. Required table columns in order: recording_id, speaker_id, start_ms, duration_ms. Supports one recording, 1–1,000 speech intervals, up to 32 speakers, a 24-hour maximum timeline, and exact millisecond timestamps. RTTM requires channel 1 and all four optional fields <NA>. TextGrid uses one IntervalTier per speaker, named RTTM:recording_id:speaker_id, with text "speech" during active intervals and empty text for gaps. Speakers may overlap across tiers, but not with themselves. Other Praat tiers, diarization metadata, transcripts, time values with more than three decimals, and nonconforming TextGrid formats are rejected. Conversions do not analyze speech or edit audio. Keep the originals and check exported annotations in the destination program.