If you already have transcripts for your video or audio files, automatic transcript alignment is the fastest and least expensive way to create captions and use interactive video plugins. This webinar covers tips and best practices for using our automatic transcript alignment service. Watch this webinar to learn about:
- How the technology works
- Uploading video/audio files
- Preparing a transcript for upload
- Transcript synchronization accuracy
- Using captions & interactive transcripts
- Editing transcripts & captions post processing
- Working with lecture capture and video platforms
Presenters:
Josh Miller (Moderator)
Co-Founder | 3Play Media
David Zylber
Manager of Customer Happiness | 3Play Media
Roger Zimmerman
VP of Research & Development | 3Play Media
2. Agenda
• Automatic Alignment vs. Transcription
and Captioning
• Alignment Service Overview
• Best Practices
• Submitting Transcripts & Media Files
• Formatting your Transcripts
• Q&A
3. Transcript Alignment Service vs.
Transcription and Captioning
• Use the Alignment service when you already
have a transcript
• Both services ultimately give you access to the
same 3Play Media account features and tools.
• Alignment is 100% automated where as the
standard service involves human clean up.
• Turnaround Service Levels
6. FTP Overview
• Create a folder named for_alignment
• Add the media file first to the for_alignment folder
- e.g. Casablanco.mp4
• Then add the plain .TXT transcript to the for_alignment folder
- e.g. Casablanco.txt
• The .TXT file MUST HAVE THE SAME NAME as the media
file
• Batch uploads: first submit all media files and then the
corresponding transcripts.
7. Alignment Best Practices
• THE KEY: Text corresponds to audio!
• Common Problems:
-Non-conforming speaker labels (not all caps, hyphens instead of colons
-Wrapped text becomes paragraphs
-Including instructions, screen directions, scene settings/headers
-Interpretation
-Overlapping speakers
-Audio quality
• Duration: No more than 2 hours per file
• Drag and Drop your transcripts when you can
• Transcripts should be unformatted plain text file (.TXT)
• Short duration reduces the likelihood of misalignment
9. Automatic Alignment Process
continued…
2) Infer verbalization from text
•
Speaker labels used for adaptation (and replaced with
optional pause)
•
Punctuation removed (sentences replaced with pause)
•
Numerics expanded:
10/10/2013 => “ten ten thirteen” OR “October tenth” …
107 => “one hundred and seven” OR “one oh seven” …
5’3” => “five foot three” or “five three” …
•
Acronyms/abbreviations expanded: “St.”, “ABC”, “NASDAQ”
10. Automatic Alignment Process
3) Build a “biased” language model (with options):
CEO: “On 10/10/2013, we will be listed on NASDAQ as ABC”
<SPEAKER> on { NULL / this } { ten ten / october tenth }
<COMMA> { NULL / twenty thirteen / thirteen } { we will / we’ll
} be listed on the nasdaq as a b c <SENTENCE> …
11. Automatic Alignment Process
4) Run ASR with biased LM:
ON
OCTOBER
TENTH
WE’LL
BE
1.02 1.05
1.05 1.32
1.32 1.51
1.63 1.76
1.76 1.82
12. Automatic Alignment Process
5) Re-Align with original text:
ON
OCTOBER
TENTH
WE’LL
BE
CEO:
On
0.0
1.02
1.02
1.05
10/10/2013,
we
will
be
1.05
1.63
1.695
1.76
1.51
1.695
1.76
1.82