Research and Critical Analysis into Audio Transcription processes
By: Kiehne, Alexander

Table of Contents

Abstract……………………………………………………………………………….3
Topic Statement………………………………………………………………………3
Work Setting………………………………………………………………………….4
Situation Analysis…………………………………………………………………….5
Problem Analysis……………………………………………………………………..7
Plan of Action………………………………………………………………………...8
Background Research 1 – Similar software…………………………………………..8
Background Research 2 – Market for software……………………………………….9
Analysis……………………………………………………………………………….11
Critical Logs…………………………………………………………………………..12 Critical log 1………………………………..12 Critical log 2………………………………..12 Critical log 3……………………………….12 Critical log 4……………………………….13
References…………………………………………………………………………….14
Appendices……………………………………………………………………………15-16

Abstract In this document the possibility of a potential loop in the market is examined. In the field of linguistic services, the process of transcription is one of the major services requested, by the legal branch of State and Federal Government Agencies. The demand for such services is high as the court system heavily relies on written documents for official use.

In time most documents transcribed could be performed by the use of sophisticated software that allows the user to accelerate the transcription process by feeding audio into a computer, allowing the software to transcribe the information. This would certainly allow for a reduction in cost as the man-hours can be reduced and the process completely outsourced. The process will take a fraction of the time and could be offered at very competitive rates.

The potential take in this industry offers incentive to pursue this idea further, as transcription will remain in high demand. Once all drawbacks are eliminated, this software would no longer create comparable current market value cost; the main focus would be on obtaining and retaining clients. Constant development and updates to the software and enterprise could be used to assure a dominating presence in the industry for a long time.

Topic Statement The art of writing down what people say has been a cornerstone of human interaction for thousands of years. The recordings our ancestors left us behind help us understand history today. The ability to record audio enables us to capture exactly what a person has said.

Transcription of documents is an important part of the court systems throughout the United States. The “Wiretap Act” (Barbour & Wright, 2009) regulates the collection of any content of wire and electronic communications. Text files are much easier to use than audio files. The legal branch has a high demand for transcription services, in order to gather background information regarding a case, and for use as evidence in court.

The ultimate goal is to create a type of software that can record English speech, automatically adapt to all accents, differentiate between speakers and transcribe the contents with correct grammar and spelling.

Work Setting A typical project from my perspective is as follows. I will receive a project proposal from a transcription agency. In the email, there will be information about the project I am being asked to work on. It will show me the length of the audio file, and the pay rate. Usually it will be by the audio minute, (i.e. $2.50/min). It will also give me a due date, by which the entire project must be completed in full. The audio file will either be attached to the email or I will receive log on clearance for the agency’s server on which I will be able to access and download the particular audio file or video. I then start listening / watching and begin to write down what the first speaker says. I will fill a table with the name of the speaker and next to it begin to type up verbatim transcription. Then I listen for the next speaker, I will then write his name into the table and write down what the person on the recording says. There can theoretically be an unlimited amount of speakers, the table below lists all speakers in the order of them speaking:

|Speaker 1: |Hello, how are you? |
|Speaker 2: |Good, thank you. |

I record their entire conversation onto a document, rewinding the audio many times to ensure the accuracy of the requested information on paper.

After finishing an assignment, I send it back via email and will confirm that the work is finished and ready to be delivered to their client. I understand that once the document is back with the agency that assigned the work they will verify completeness and check the document in full to ensure quality before final delivery to the client.

When working on transcriptions, I use audio programs (I.e. VLC player, Express Scribe) to slow down the audio/ enhance the sound quality. These programs work well with hotkeys. Being able to have access to these programs is a clear help to the industry of linguists. They strongly rely software that is both easy to use but efficient for their field of work. Some programs even offer foot pedals to aid a faster pace of work as you can stop or slow down/ speed up the audio with your feet leaving your hands ready to type.

Situation Analysis To explore the possibility of voice recognition software that would potentially transcribe audio and thereby make the entire process more efficient, i.e. cheaper. It seems archaic that we have yet to develop such a program, when cars, telephones, and even Google are voice controllable at this point. Most projects will be broken up into pieces of about an hour in length and sent to different linguists that will work simultaneously in order to meet tight deadlines. The quality of the transcription must be very high as these documents are often being used in court proceedings and people’s lives may depend on it. Each linguist needs to be as accurate as possible if not perfect. The different languages and dialects of the world are too many to develop such a program for each and every one. I do believe that it would be possible to create a program that is flawless at least for the transcription of English audio.

The greatest threat to this idea is to adapt to different speakers, programs have to have clear instructions so far and adapt to the speaker only when used by the same person. Thus creating the chance for the program to adapt to the speaker and become a great asset in the medical industry where dictation is common to let doctors record their ideas. The language services industry is in need of a program with the capacity to differentiate between speakers. Being able to adapt to accents and still be 100% precise in grammar and content is definitely a hurdle to overcome.

In a perfect scenario all people would speak flawless English, recordings would be clear and speakers would wait their turn, but this is usually not the case. It is challenging work to transcribe audio; this work should not be underestimated. It can get confusing to differentiate between speakers. Sentence structure of verbal English is very different from written English, and to write down exactly what people say is time consuming work.

Problem Analysis Transcription is based on writing down exactly (word by word) what people say. This can be tedious work, depending on the medium used for transcription. In my work I have come across many recordings, for example, wiretaps, prison phone calls and even videos of interrogations during the legal process. Besides the fact that the content of these recordings may be quite interesting, transcription has some flaws. It is very time consuming, it simply takes time to listen to the recording, write it down, go back and forth, check the work and finally be done. This also costs a lot of money for the Agencies that offer these services to branches of law-enforcement.

In addition, if all agencies that offer these language services worked together on finding a way to accelerate the process, they could split the cost of developing it and maintain competition. Having such a program would immediately prove lucrative as you no longer have to seek out linguists, work out terms for an assignment, work out pay rates, wait for the assignment to be completed, and then check it again once it is back and completed with the agency.

I think the major challenge with such a program would be the fact that it would have to be sensitive to accents and dialects and the difference in meaning of the content due to the changes of grammar. I believe that this could be worked out if the programs were open source and flaws would be eliminated faster that way.

Plan of Action The potential in this industry has not been exploited; because of the economic downturn in addition to the increasing rate of unemployment crime will remain on the rise (see Appendix A). The US Economy has not been the economic powerhouse it used to be, the credit crisis has been forcing the US government to borrow money when there is not enough, resulting in a steadily growing national debt since the 1970's, and make dealing with the country's long-term debt and deficit more difficult. This is relevant because economic trends are closely linked to crime; The US economy has an immediate effect on this industry. If crime goes up, the requests for transcription services rise accordingly. In a country like the United States of America, where legal wiretaps are commonly used to bring criminals to justice, it could be an opportunity to create a program that would reduce human workload. In an ideal world, the program would completely eliminate the need for human labor but I don’t think that is a realistic option given the current state of software capability.

In the USA alone, if we were to offer the industry perfect transcription for all criminal cases that used wiretaps to seek convictions, it would be a service needed by most courts. Once developed, it would certainly create revenue as the cost of human labor could potentially be almost eliminated.

Background Research 1 / Similar software There are several similar programs in the market today, however these programs have issues differentiating between similar languages, and only work at an acceptable level when the recordings are extremely clear. This is an issue because some recordings are obtained in public and at times background or travel noise could be a factor. With the aid of high-end software these noise levels can be reduced, and brought down to a level that still allows the linguist to capture the content of the conversation (speech), but programs that have been developed still have major problems and cant be used efficiently yet.

The most basic program of voice recognition fetches about $200 on the internet (Amazon, ebay, etc.); Nuance Naturally speaking, Dragon Dictation, but those only lets you control/ interact with your computer/ phone and helps during text entry but does not follow grammar rules unless prompted (”full stop”; “line break”; “next paragraph”) and still has difficulty with accents and cannot differentiate between speakers. The most important aspect of the program needs to be a reliable source of transcribed information. Express Scribe software is designed for human transcription services and offers features such slowing down playback and enhancing the desired audio. It allows the linguist to work faster as the audio is clarified with the software. The military actually frequently uses a program called “SpeechTrans”, that allows soldiers to easily translate into many different languages. (MultiLingual, July/ August 2011; John Freivalds, Winning the war on speech apps)

Background Research 2 / Market for software Crime is a constant part of our society; the main client of this industry is the legal branch of the federal government and state governments. The use of wiretaps and other listening devices are common techniques used to bring criminals to justice by the departments of law. Because of the success of successful convictions by the effective use of wiretaps the need for the transcriptions of those wiretaps will remain.

The success of wiretaps is evident, as we all know about its use in law-enforcement. The success rate for obtaining arrest warrants using wiretaps is 100% and about 50% of all criminal cases in the US using wiretaps lead to convictions. (see Appendix B)

In recent months Raj Rajaratnam and Zvi Goffer have been convicted with the aid of wiretaps. In white color crime, the exchange of information is the most valuable commodity, convictions for i.e. insider trading are rare, but wiretaps with clearly recorded exchanges of information on classified trading secrets are giving the attorneys a better chance to prove charges.

” Federal prosecutors in New York disclosed in court papers that they had a court-authorized wiretap on his mobile phone in connection with their case against Donald Longueuil, a former SAC Capital Advisors LP portfolio manager who pleaded guilty to insider- trading charges on April 28 (2011) in Manhattan federal court.” (BloombergNews.com)

Transcription is one of the fastest growing legitimate jobs in the US and there is plenty of room for growth as the need for transcription services is constantly growing.

“In the year 2006, around 14 million transcribers were working in the United States and it is expected to grow by another 3 million before 2015” (http://www.transcription-connection.com/)
Analysis
After studying the subject of transcription and examining the process of transcribing in the light of offering a new service to this industry, I believe there is a market for the suggested software.

Such software would certainly be of high demand for all courts and would eliminate some human jobs. Cutting cost is a constant priority in business, it is projected that by 2026 computers will overtake humans in terms of capability of computing. The interaction with electronic devices has become standard in our lifes, to the point that every car has voice controlled features, every computer can understand vocal commands and cellular communication has reached the point where videoconferences are possible from your phone with any given person regardless of location, worldwide.

With the aid of funded development it would be possible to explore the possibility further. Programming experts would be needed to assure guidance of the scientific aspect of the development process. I am convinced that with the projected advances in tech sector, this will be a bubble that could dominate this part of the industry and potentially eliminate most competition.

Instead offering the same services as other agencies in the language services industry, the possibility of offering the same service, without fault or fail, in a fraction of the usual timeframe whilst still charging the lowest rates in the market. With fully functional software the potential for revenue is high. The potential investment for starting this company would be minor when compared to the potential money to be made. The US Attorney’s Office alone spends several millions on a monthly basis for transcription services. There is constant demand for this service and there will be for years to come.

Critical Log 1 Classified documents and privacy issues regarding ownership of the recorded and documented property; Google voice for example, grows (gets smarter) because it saves all entry information growing its library of content allowing it to recognize concepts faster, thus becoming a more user friendly tool. The Government would have to allow this program access to all files and compare audio and text to learn to recognize “active speech vocab”, slang and incorrect grammar that holds meaning in verbal English.

Critical Log 2 Elimination of human component is not 100% possible. Ideally, you would feed audio information into a computer and get a print out with the information in text. Would it be possible to assure 100% quality so that it would not need to be reviewed by a person? Theoretically, yes. But I think realistically, no. I think the accents and content of meaning prevent this possibility from being realistic given the current state of the art of software standards for transcription.

Critical Log 3 Language service agencies would be eliminated with such software or would have to be reorganized entirely as the landscape of the industry would change…Federal agencies might be interested in investing for this program as opposed to outsourcing to 3rd party agencies.

Critical Log 4 Discovered a program online called AppTek that supports numerous languages, as well as multiple speakers, and it provides full transcription. This essentially is the program I suggested offered with most services possible from a technological standpoint. They offer a free demo, which I tried out on an older work assignment of mine and then compared it to my transcription. I found that, the program works slow and quit unexpectedly 4 different times while transcribing 30 minutes of audio. I also came to the conclusion that it has difficulty with regional accents, i.e. the new york accent for the word “all” sounds similar to the word “hall”. This was a reoccurring mistake. Similar sounding words seem to create problems for the AppTek transcription program. All in all however I think this is the early stage of a program that I thought was not yet available.

References

Motivans, M. (2010) Trends in the types of cases concluded in U.S. district court: Trends in the types of cases concluded in U.S. district court. Federal Justice Statistics, 2008 – Statistical Tables. Retrieved from http://bjs.ojp.usdoj.gov/content/glance/tables/fedtyptab.cfm

1. Freivalds, J. (July/ August 2011) Winning the war on speech apps. MultiLingual.

2. Hurtado, P. (2011) Expert-Network Founder Who Refused FBI Wire Expects Arrest. Bloomberg News. Retrieved from http://www.bloomberg.com/news/2011-07-08/expert-network-founder-who-refused-fbi-wire-says-he-expects-his-own-arrest.html

3. Barbour, C. & Wright, G. (2009) Keeping the Republic: Power and Citizenship in American Politics. Washington, DC: CQ Press

4. http://www.transcription-connection.com/

5. http://www.landmarktranscription.com/why-landmark.html

6. http://www.verbalink.com/

7. http://www.apptek.com/

Appendix A

[pic]

Source: Federal Justice Statistics, 2008 – Statistical Tables.

Appendix B
[pic]