Xpio Health · Library of Congress RFI 2027-NLS-0075 · Pilot demonstration

Automated talking book production, built to NLS Specification 1203

Two complete digital talking books produced with no manual step: one from a commercial audiobook with synthetic NLS announcements, one narrated entirely by synthetic speech from an EPUB. Both target ANSI/NISO Z39.86-2002, the version Spec 1203 requires.

Listen and navigate

Navigation (NCX)

"Hear label" plays the heading clip from the headings file, the way an NLS player voices navigation.

Ready
Choose a heading or press Play.
#clipBeginclipEndLength

Each row is one SMIL clip. Playback follows the clips, the way a DTB player does: silence between clips is skipped. Amber rows are headings. This page streams a preview of the first 2½ minutes of each audio file; the full book is in the delivered file set.

Conformance report

ResultCheckClause

What the pilot covers

  • Both SOW production paths: commercial audiobooks (3.1) and full-length synthetic speech from text (3.2).
  • Built to Z39.86-2002 with the 2002 DTDs and OEB 1.0.1 package format that Spec 1203 cites, shipped inside each book.
  • Clip timing measured from the master audio: every clipBegin 80–120 ms before narration and every clipEnd 150–300 ms after.
  • Audio compressed to AMR-WB+ (frame type 23, ISF index 8) in 3GP with the 3GPP reference encoder, including the md5sum keyword and constant-size sample table Spec 1203 §3.2.3 requires. Decoding it back measures a codec offset of 66–68 ms, inside the 50–80 ms window in §3.2.5.
  • Independent check with the DAISY Consortium's ZedVal 2.1 validator: no failures in the 18 of its 21 test groups that run on current Java. Its only notices on the delivered books are that it predates NLS's 3GP audio format.
  • SMIL files packed to the 100 KiB limit, MD5 checksum file, headings file, opening and closing announcements.
  • Pluggable voices: a local open-source engine (no per-use fee, no content leaves the system) or Amazon Polly in the Library's AWS account.
  • The Library owns all source code. No subscription or license is required to run it.

Not yet shown

  • Playback on NLS hardware players. Validation here is automated and in the browser.
  • PDTB2 encryption (Spec 1205). It needs Library-held keys; we recommend a Government-controlled key interface.
  • Deployment into the NLS AWS environment. This demonstration ran on one workstation; the pipeline is a command-line tool that can run as a batch job on AWS.
  • Page and note navigation (navLists). The sample books have no print page numbers.