--- title: "MACROPHONE: An American English Telephone Speech Corpus" description: "The paper describes Macrophone, a large corpus of American English telephone speech for training and… 5 references, cited by 2 papers here." authors: ["Kelsey Taussig", "Jared Bernstein"] year: 1994 venue: "Human Language Technology: Proceedings of a Workshop held at Plainsboro, New Jersey, March 8-11, 1994" source: "https://aclanthology.org/H94-1007/" licence: "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)" licence_url: "https://creativecommons.org/licenses/by-nc-sa/3.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/taussig-1994-macrophone-american-english-telephone-speech/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/taussig-1994-macrophone-american-english-telephone-speech/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [1994](https://md.research.elamur.ai/free/en/papers/1994/) / MACROPHONE: An American English Telephone Speech Corpus # MACROPHONE: An American English Telephone Speech Corpus **Authors:** Kelsey Taussig and [Jared Bernstein](https://md.research.elamur.ai/free/en/authors/jared-bernstein/) · **Year:** 1994 · **Venue:** [Human Language Technology: Proceedings of a Workshop held at Plainsboro, New Jersey, March 8-11, 1994](https://md.research.elamur.ai/free/en/venues/human-language-technology-proceedings-of-a-workshop-held-at-plainsboro-new/) ([Human Language Technology: Proceedings of a Workshop held at Plainsboro, New Jersey, March 8-11 1994](https://md.research.elamur.ai/free/en/venues/human-language-technology-proceedings-of-a-workshop-held-at-plainsboro-new/1994/)) · **Licence:** [CC BY-NC-SA 3.0 (ACL Anthology, before 2016)](https://creativecommons.org/licenses/by-nc-sa/3.0/) **Links:** [Publisher's page](https://aclanthology.org/H94-1007/) · [PDF](https://aclanthology.org/H94-1007.pdf) · [OpenAlex](https://openalex.org/W2027610368) **Citations:** 6 citations (OpenAlex, 2026-10-14) · 2 in our library, of the 149,439 papers whose bibliographies we read ## Summary The paper describes Macrophone, a large corpus of American English telephone speech for training and evaluating recognizers in automated phone services such as ticketing, scheduling, banking, and database lookup. The Linguistic Data Consortium (LDC) sponsored it as the first of a planned multilingual series called Polyphone. About 5,000 callers, recruited by mail through a market research panel, dialed a toll-free number and followed a unique printed prompting sheet. Each sheet mixed read items (digit strings, money amounts, dates, times, names, addresses, spelled words, and TIMIT/WSJ/ATIS sentences) with spontaneous answers to spoken questions. Speech was captured digitally from T1 lines as 8-bit mu-law. Temporary workers then checked each file against a default transcription or transcribed it, and passed hard cases to a linguist. The outcome was roughly 200,000 verified utterance files with demographic SPHERE headers, produced in five calendar months once the material was fixed. The project shows that a corpus of this size can be planned and delivered on a predictable budget of time and labor. ## Key points - Each sheet asked for 45 responses: 34 read and 11 spontaneous. The spontaneous part had 10 fixed prompts, mostly yes/no or demographic, plus 1 prompt rotated from a small set. - The prompt pools were built together: - place names: North American cities over 150,000 people and world cities over 2 million; - personal names: 600 first names and 600 surnames, each set chosen by frequency of use; - street names: 971 names taken from the ZIP+4 directory; - application words: 674 control words. - Out of a panel of 400,000 US households, 20,000 sheets were mailed in six waves. They produced 6,700 calls, a 33% response rate, against a planned 25%. - An average call lasted about six minutes and gave about four minutes of speech, of which roughly a third was silence. About 7.5% of files were discarded as empty or truncated. - The hardware was Dialogic boards in a 386 PC with 10 lines, feeding a Sun SPARCstation. Files carry SPHERE headers and were sent to LDC on Exabyte tape for CD-ROM release. ## Interesting ideas - Each prompt pool was designed with the others in mind, so an item in one pool was kept out of the others (a surname would not reappear as a street name) and every category stays unambiguous. - Because every sheet was unique and carried an ID, workers could load an expected transcription and only confirm or correct it. Verifying a default is much cheaper than transcribing from scratch. - Utterances with fragments, mispronunciations or other disfluencies were split off and sent to a linguist, while routine files went through the cheap path. - The mailing sample was deliberately skewed to offset known differences in response rates by sex, age and income, so the corpus that came back would be balanced. - The structured read categories (dates, money, times, fractions, addresses, units) were chosen to match exactly what target applications must understand. ## For Mova How these ideas could serve Mova, our small open language model — its files, stages and tests: - **I1** (world): When compiling gazetteers or name lists into the ontology, cross-checking lists against each other could flag entries whose type is ambiguous instead of silently giving them two types. - **I2** (learn): Silver parses could be shown to the LLM teacher or a human as a "default to confirm or fix" rather than produced cold, saving annotation effort. - **I3** (learn): When witness parsers disagree, the sentence could be escalated to a higher teacher tier, and agreed sentences stay in the low-cost pipeline. - **I4** (learn): Evaluation and training sets could oversample text types where the parsers are known to fail more often, so the final sample still reflects target coverage. - **I5** (tokens): A matching test suite of such entity formats could stress the tokenizer and measure normalization, for example "$1,975.55", "a quarter to twelve", or "4.32 grams". *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* ## Institutions - [SRI International](https://md.research.elamur.ai/free/en/labs/sri-international/) — Research institute, Menlo Park, United States · 24 works · 12,175 citations 📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works 🧭 Topics (2): [English](https://md.research.elamur.ai/free/en/topics/english/) · 2,379 works · [🎙️ Speech & spoken language](https://md.research.elamur.ai/free/en/topics/speech/) · 519 works ## Cite Cite as: Kelsey Taussig and Jared Bernstein. 1994. MACROPHONE: An American English Telephone Speech Corpus. Human Language Technology: Proceedings of a Workshop held at Plainsboro, New Jersey, March 8-11, 1994. The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{taussig1994macrophone, title = {{MACROPHONE: An American English Telephone Speech Corpus}}, author = {Kelsey Taussig and Jared Bernstein}, booktitle = {Human Language Technology: Proceedings of a Workshop held at Plainsboro, New Jersey, March 8-11, 1994}, year = {1994}, doi = {10.3115/1075812.1075820}, url = {https://doi.org/10.3115/1075812.1075820} } ``` - [← previous A Freely Available Syntactic Lexicon for English](https://md.research.elamur.ai/free/en/papers/egedi-1994-freely-available-syntactic-lexicon-english/) - [next → Computational Analyses of Arabic Morphology](https://md.research.elamur.ai/free/en/papers/kiraz-1994-computational-analyses-arabic-morphology/) - [↑ back to papers of 1994](https://md.research.elamur.ai/free/en/papers/1994/) ## References 5 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. 1. L . H i r s c h m a n - M A D C O W (1992): "Mu, lti-Site Data Collection for a Spoken Language Corpus, ' Prec. DARPA Speech and Natural Language Workshbp, M.Mareus (ed.), Morgan Kaufman. pp. 7-I4. 2. [find](https://scholar.google.com/scholar?q=1%2E+L+%2E+H+i+r+s+c+h+m+a+n+%2D+M+A+D+C+O+W+%281992%29%3A+%22Mu%2C+lti%2DSite+Data+Collection+for+a+Spoken+Language+Corpus%2C+%27+Prec%2E+DARPA+Speech+and+Natural+Language+Workshbp%2C+M%2EMareus+%28ed%2E%29%2C+Morgan+Kaufman%2E+pp%2E+7%2DI4%2E) 2. L.Lamel, R. Kassel & S.Seneff (1986): "Speech Database Development: Design and Analysis or me Acousfic-rnonetie Corpus," Prec. [find](https://scholar.google.com/scholar?q=L%2ELamel%2C+R%2E+Kassel+%26+S%2ESeneff+%281986%29%3A+%22Speech+Database+Development%3A+Design+and+Analysis+or+me+Acousfic%2Drnonetie+Corpus%2C%22+Prec%2E) 3. DARPA Speech Recognition Worxsnop, February 1986, pp. 100-109. 3. [find](https://scholar.google.com/scholar?q=DARPA+Speech+Recognition+Worxsnop%2C+February+1986%2C+pp%2E+100%2D109%2E+3%2E) 4. D.Paul & J.Baker (1992): "T,h,e Design for the Wall Street Journal-based CSR Corpus,' Prec. DARPA Speech and Natural Language Workshop, M.Mareus (ed.), Morgan Kaufman. pp. 35T-362. 4. [find](https://scholar.google.com/scholar?q=D%2EPaul+%26+J%2EBaker+%281992%29%3A+%22T%2Ch%2Ce+Design+for+the+Wall+Street+Journal%2Dbased+CSR+Corpus%2C%27+Prec%2E+DARPA+Speech+and+Natural+Language+Workshop%2C+M%2EMareus+%28ed%2E%29%2C+Morgan+Kaufman%2E+pp%2E+35T%2D362%2E+4%2E) 5. B.Wheafley & J.Pieone (1991) "Voice Across America," Digital Signal Processing, 1, pp. 45-63. [find](https://scholar.google.com/scholar?q=B%2EWheafley+%26+J%2EPieone+%281991%29+%22Voice+Across+America%2C%22+Digital+Signal+Processing%2C+1%2C+pp%2E+45%2D63%2E) *Updated 31 October 2026*