Transcript editor

#This application is related to U.S. patent application Ser. No. 12/748,695 entitled “Spot Dialog Editor,” filed on even date herewith, which is wholly incorporated herein by reference.

#Interviews and most documentaries are often recorded without the prior preparation of a script, or even of a detailed story board. Frequently, an interviewer, sound engineer, or camera operator simply turns on a microphone or starts shooting with a video camera while the subject is being interviewed. After the capture is complete, the raw media needs to be edited, often resulting in a program containing only a small fraction of the material originally captured. Typically, the editing workflow proceeds as follows. First a transcript of the captured media is prepared by hand. Then a producer or a story editor marks up a printout of the transcript to indicate the sections of the program to be retained in the story. Next, an editor receives the marked up transcript printout, and uses it as a guide to locate the media corresponding to the portions to be retained, loads these media portions, plays it, finds the media portions of interest, and then performs the edit. This process is laborious and time-consuming, in part because of the constant need to navigate back and forth through printed pages and the original media (video and/or audio), and to wait till the media is loaded and playing before performing the edits.

#The methods, computer program products, and systems described herein involve generating a rough cut of a time-based media program that includes recorded dialog by editing a transcript of the media without referring to the media itself.

#FIG. 1 is flow diagram for generating a script with media timing information.

#FIG. 2 is a flow diagram for generating a rough cut of a media program by editing a transcript.

#FIG. 3 shows a graphical user interface for a transcript editor.

#In a dialog-intensive video or audio program, such as an interview, the major portion of the program content is contained in what is said during the program, rather than in the visual information, which is typically a shot of a person talking, often interspersed with background material, which is referred to as B-roll. Since text is easier and faster to read and edit than video or audio, most documentaries and interviews are edited “on paper” first, to generate what is referred to as a “paper cut” or “radio cut.” Thus, the current workflow for creating such programs starts with the original footage, which is transcribed by hand and marked up with timing information indicating the location of the corresponding footage in the original media, and the character who is speaking. A producer or story editor annotates the transcript to indicate which parts of the program are to be retained, and in what order. The editor then uses the annotated transcript, and its associated timing information, to find the clips within the original footage, and edits them to produce the finished program. The process involves two labor-intensive, manual processes. First, the transcriber must listen to all the original footage in order to transcribe the audio, noting the location within the original footage, such as time code, offset, or other time or frame indicator, and the identity of the character speaking. Second, the editor refers to the annotated transcript in order to locate the media to be edited, and builds up a timeline of the program being edited by referring back and forth between the annotated transcript, and the media.

#In the workflow described herein, the process of creating a rough cut edit of a speech-intensive program is greatly accelerated. The editing methods described can be used on any media that includes audio speech, and specifically apply to video accompanied by audio, or to audio only. As used herein, dialog refers to spoken speech, including speech spoken by one or more people.

#The workflow starts with a transcript that has been created by a hand-transcriber playing back the original media. In generating a word-accurate transcript, the transcriber may be assisted by using a tool such as the spot dialog editor, described in related U.S. patent application, entitled “Spot Dialog Editor, Ser. No. 12/748,695. In the first stage of the process, the transcript is augmented with word or phrase-accurate timing information. In the second stage, the augmented transcript is edited in the absence of its corresponding original media to produce an output file that can be input into a media editing system, such as video editor or digital audio workstation, to automatically produce a rough cut of the edited program. FIG. 1 is a flow diagram showing the main steps in the first stage. The system loads audio dialog track 102 of the media to be edited into an audio recognition system that analyzes the speech in the audio dialog track and identifies the phonemes present within the speech (step 104). This process generates phoneme audio track 106, which contains a sequence of the identified phonemes and the time codes of the locations within the audio dialog track where they occur. Note that it is not necessary to recognize the words or understand the text being spoken—detection of the phonemes is sufficient. Such an engine is available from vendors such as Nexidia Inc. of Atlanta Ga., which offers products that can process audio at rates of 250 or greater times faster than real time. For each program being edited, the system also has access to transcript 108. Breaking text into text-based phonemes (step 110) is readily accomplished, to generate the phoneme version of the transcript (112). The system then compares phoneme-based transcript 112 with phoneme audio track 106, attempting to match up the phonemes in the transcript with those in the audio track (step 114). When it finds a match, the system inserts into the transcript the timing information from the matching word or phrase in the phoneme audio track. The timing information is expressed as a timing offset from the head of the track; alternative representations of the timing information include a time code and/or frame number. In sum, the media essence is indexed by phoneme, is associated with metadata (timing data), and is inserted into the transcript. One system for producing such a time-augmented transcript is described in U.S. Pat. Nos. 7,231,351, and 7,263,484, which are wholly incorporated herein by reference. Other methods of generating the timing information and adding them to the transcript may be employed, such as for example speech recognition, with matching at the word level rather than at the phoneme level. The methods described herein are not dependent on the specific manner in which the timing information that is used to augment the transcript is obtained.