and it didn't seem to have a log of words labeled each by timestamp offset into the audio recording - which is the vital part for training a recognizer. Am I missing something?
Bottom line: speech recognition in the general case (more than a few predetermined words) is only as good as the 1) acoustic model (which utterances were heard), and 2) language model (how do we group the utterances into words).
This requires massive amounts of labeled data. This is why Nuance is king and few others come close - the amount of labeled data necessary to catch up is astounding. Not to mention a patent minefield to navigate.
This is unfortunately one field in which open-source alternatives face real obstacles and won't be viable in the near future.
Google already does not fulfill subpoenas for email seizures in many cases due to the decision in Warshak - their contention is that they cannot be sure if the user resides or uses their email under the jurisdiction of the Sixth Circuit.
It's puzzling to think about why Warshak wasn't appealed to the Supreme Court (SC), where a decision would have federal instead of regional consequences. The Justice Department most likely felt such a move could easily end in the SC siding with the Sixth, ending the free lunch on electronic communications seizures that law enforcement currently enjoys.
This is the most informed post here so far. A couple of important points to add.
The problem is 2703(f). It allows a government entity to request a snapshot of a user's account "pending the issuance of a court order or other process", meaning that 180 days later, the government can just issue a subpoena - they get their foot in the door and then come back later with the lowered bar, and only need a subpoena. Same effect as not having warrant protection.
http://www.law.cornell.edu/uscode/text/18/2703
It's important to point out that the 90-day notification can be renewed indefinitely, pursuant to 2703(f)(2), which is probably a breach of what the 4th Amendment intended.
I downloaded a couple here:
http://www.repository.voxforge1.org/downloads/SpeechCorpus/T...
and it didn't seem to have a log of words labeled each by timestamp offset into the audio recording - which is the vital part for training a recognizer. Am I missing something?