I am looking for a software to support phonetic transcription of a sound recording in a phonetically undocumented language. I have heard about PRAAT, but I wonder if there is a software that can identify isolated recorded sounds according to the IPA.
I've used Transcriber, which is now TranscriberAG. It's a really useful program for transcribing texts and gives you tools for slowing the tempo of the sound file, as well as annotating sections of the sound file itself.
Praat was still necessary for difficult parts of recordings where spectrograph analysis is essential.
Regarding automatic annotation, the technology just isn't here yet - as far as I know. Automatic speech-to-text technologies are improving rapidly, but only in big-data fields with resources as large as e.g., Google, Apple, etc. So automatic identification of extremely small segments of speech, in sometimes low-quality recordings, from a wide range of speakers, is still perhaps years away. So you're likely better off training your ear or hiring an experienced phonetician.
I recommend searching YouTube for Praat tutorials, and starting off with Ladefoged's "A Course in Phonetics."
http://trans.sourceforge.net/en/presentation.php
http://transag.sourceforge.net/
Book A Course in Phonetics
Phonetic transcription is a complex task involving not only objective decisions about the acoustic signal, but (depending on the transcriber) such things as inferences about unseen articulatory patterns and relations between sounds in a particular language's phonology.
I am not aware of any software that even tries to automatically produce phonetic transcriptions from an acoustic recording.
The reason people (like me) point to Praat is because it allows you to examine the sound in ways that help a phonetically-trained listener most efficiently produce a phonetic transcription. You can listen to any portion of the sound you choose - a single phone, a transition, a whole word, the entire utterance, etc. You can also see the waveform and spectrogram, and request information on formants, amplitude, frequency, etc.
I know this isn't what you're after, but hopefully it clarifies why it will be hard to find the software you are describing.
If such software existed, I would love to know about it. At the least, it could be a valuable tool to further aid learning and production of phonetic transcription by humans. At best, it could overcome the difficulties and serve as a replacement for human transcribers in certain situations.
Hi Stefan,
You could use Praat indeed. Mart Lubbers and Francisco Torreira published a tool called praatalign on Github that helps segmenting different languages. If you had sound models for a related language that HTK could use one could try segmenting your sound files after an orthographic annotation. I guess that you want to do work on a Quechua related language? Maybe Simeon Floyd (ATmpi.nl) might know about ASR accounts for Quechua. It would be interesting to know what you plan to do with the annotations. That could affect the analysis approach.
I've used Transcriber, which is now TranscriberAG. It's a really useful program for transcribing texts and gives you tools for slowing the tempo of the sound file, as well as annotating sections of the sound file itself.
Praat was still necessary for difficult parts of recordings where spectrograph analysis is essential.
Regarding automatic annotation, the technology just isn't here yet - as far as I know. Automatic speech-to-text technologies are improving rapidly, but only in big-data fields with resources as large as e.g., Google, Apple, etc. So automatic identification of extremely small segments of speech, in sometimes low-quality recordings, from a wide range of speakers, is still perhaps years away. So you're likely better off training your ear or hiring an experienced phonetician.
I recommend searching YouTube for Praat tutorials, and starting off with Ladefoged's "A Course in Phonetics."
http://trans.sourceforge.net/en/presentation.php
http://transag.sourceforge.net/
Book A Course in Phonetics
Other tools for forced phonetic alignment include: EasyAlign, by J.-P. Goldman, and Train&Align, by S. Brognaux et al.
First you have to do transcription using "Wavesurfer" speech analysis software and IPA chart. After that you can implement a phonetic Engine using HTK software. Phonetic Engine is a system which can do automatic transcription of the speech test utterances.
I have used Praat but principally for spectral analysis and measurements. Your case is pretty difficult because the language you are working on may not be documented and therefore the sounds (phonemes) involved may not be IPA compatibles. Besides, even if you find a software that does it, please take the warning from Ladefoged who asserted that instruments are only a help, but the final arbiter in all phonetic judgement is the trained ear of a phonetician. Regardless of the size of the data you have to transcribe, you will find that you are assured of the best results doing it manually. Even if you use a software, you'll still need to listen and ensure that the software had done its job. But if you are working on a language whose phonemic units you are not conversant with,, I would be scared to think that you can get a machine to do the work. Good luck. .
I know I am a bit late to add to this thread, but I also wanted to mention a combination of two interoperable tools that we have developed at our lab (shameless plug alert!!):
WebMAUS Multiple which is part of the BASwebServices:
https://clarin.phonetik.uni-muenchen.de/BASWebServices/#/services/WebMAUSMultiple
can be used to generate forced aligned phonetic corpora using MAUS in the emuDB format. This format can then be used by the EMU-SDMS system:
http://ips-lmu.github.io/EMU.html
which includes a web based transcription / annotation tool called the EMU-webApp:
http://ips-lmu.github.io/EMU-webApp/
to inspect & manually correct the forced aligned annotations.
- Badges
- Science topic
- Similar topics
- Linguistics