Background
Adobe Premiere has a speech-to-text translator, as part of its content-analysis capability. At best it is 80% or so correct in its interpretations, though in my experience only 20-30% reliable. But to optimize its chances, one must select the (spoken) language appropriate to the media (content) being analyzed. But by default, only one language, US-English is available. So how do you get further options?
Summary:
- By default, the only language model (sic) installed is that for US-English.
- Optionally, one can download (free) Installers for other language modules.
- One can download the installer for International English language models (sic), from http://www.adobe.com/uk/products/premiere/extend.displayTab3.html
- These English-language models include: Australian, British, Canadian.
- Run the Installer
- Although intended for both CCand CS6, it only installs to [C:\Program Files (x86)\Common Files\Adobe\SpeechAnalysisModels\CS6]
- Manually copy content from [C:\Program Files (x86)\Common Files\Adobe\SpeechAnalysisModels\CS6]
to [C:\Program Files (x86)\Common Files\Adobe\SpeechAnalysisModels\4.0]- (sic)
- Likewise, for Mac OS:
- Copy all content of [/Library/Application Support/Adobe/SpeechAnalysisModels/CS6
- to [/Library/Application Support/Adobe/SpeechAnalysisModels/4.0]
- Incidentally, it is possible to inject (eg via C++ code) a text script directly into XMP metadata
- See Details for a link and example code.
Details:
Lacking a script, I wanted to use Adobe’s Speech-to-Text function in Premiere’s Analyze Content feature. Realizing that such analysis is actually performed by Adobe Media Encoder (AME), its results (typically) being appended to given media file in the form of XMP data, I opened AME, then its Help page then searched on [
For the UK:
- http://www.adobe.com/uk/products/premiere/extend.displayTab3.html
- Speech Analysis Models for Adobe Premiere Pro CC and CS6
- For Windows
- Adobe Speech Analysis Model — American English 139MB
- {I guess it comes bundled with this one ???}
- Adobe Speech Analysis Model — International English 285MB
- {I guess this is the nearest thing to British English ???}
- {Ought really to have more kinds of English…}
- Adobe Speech Analysis Model — American English 139MB
I downloaded and ran the “model”. It installed to:
- C:\Program Files (x86)\Common Files\Adobe\SpeechAnalysisModels\CS6
- That location initially contained only [en_US] and [support], the latter containing [AmericanEnglish] (among other folders).
However on re-starting Premiere CC, this language module was not apparent.
Someone at [http://premiere62.rssing.com/chan-7308938/all_p251.html] encountered the same problem, even tried creating an equivalent “CC” folder containing a copy of the content, but no joy…
Solution was at: http://forums.adobe.com/message/5477813#5477813.
- put all content of [C:\Program Files (x86)\Common Files\Adobe\SpeechAnalysisModels\CS6]
- {I took “put” to mean “copy”}
- in -> [C:\Program Files (x86)\Common Files\Adobe\SpeechAnalysisModels\4.0]
- that’s all.
This worked.
Likewise, for Mac OS:
- Put all content of [/Library/Application Support/Adobe/SpeechAnalysisModels/CS6]
- in [/Library/Application Support/Adobe/SpeechAnalysisModels/4.0]
Incidentally, one user developed C++ code as follows, to inject a text script directly into XMP metadata.
- http://forums.adobe.com/message/5202092
- This assumes script in a text file, and bypasses the normal Adobe route of going via Adobe Story.
- Code:
- cout << “Retrieving metadatas…” << endl;
- file.GetXMP(&meta);
- cout << “Injecting transcription into metadata…” << endl;
- meta.AppendArrayItem(kXMP_NS_Script, “dialogSequence”, kXMP_PropValueIsArray, NULL, kXMP_PropValueIsStruct);
- meta.SetProperty(kXMP_NS_Script, “dialogSequence[1]/xmpScript:character”, “SPRECHER”, NULL);
- meta.SetProperty(kXMP_NS_Script, “dialogSequence[1]/xmpScript:dialog”, transcription, NULL);
- cout << “Writing metadatas…” << endl;
- file.PutXMP(meta);