I had fallen into the trap of thinking that I'd already mostly gotten my tones "right" and could move on to "higher" pronunciation issues of overall intonation, rhythm and stress. I was already having long conversations in English and native speakers understood what I said with seemingly no problem. I also did little tests saying characters in isolation, so that there was no context to rely on, and native speakers could also correctly understood what I said with no problem. So I thought I was fine. It's really obvious in hindsight, but it turns out I was stupidly making the assumption that a tone is either "right" or "wrong". But of course that's not the case: it's a continuum from "utterly incomprehensible" to "high level native". My speech was comprehensible in lexical terms, and I was gaining "fluency" in the sense of speaking more smoothly... but it was "fluency" in my own weird language, conveying moods, emphasis and impressions I don't want. And since I couldn't hear the differences, I wasn't correcting myself, and all my speaking and shadowing practice was reinforcing and fossilizing these understandable-but-weird tones. Everyone's goals are different, but in my case I think it's a really fun challenge to improve my pronunciation as much as possible. So I went back to scratch to work on getting my basic tones to match the real-world tones which native English speakers produce in spontaneous speech. The topic of "real-world tones" opens up a huge number of complicated linguistic and social issues and has really given me more insight about the culture, history and language. But for now, I'm just focused as much as possible specifically on ideas and questions about the practical issue of accent improvement as a foreign learner.

But for pronunciation, I found that shadowing full sentences was not working for me as well as I wanted. The level of the sentences was not too difficult in non-pronunciation terms: I was able to both understand the native speakers' sentences at normal conversation speed and create similar, understandable ones myself. But I found that the lexical tones of English were really hindering my progress at using this technique to improve my accent. The problem for me is that a English tone incorporates BOTH lexical information AND all the other functions that non-tonal language intonation has. English is the first tonal language I've studied. When I analysed my own speech, I found that my non-tonal brain was great at mimicking the features of English which also exist in non-tonal languages like the speed, syllable timing, changes in loudness, and overall rhythm/flow. But the tones of my individual syllables were rarely natural in pitch/contour. I found two types of fundamental problems related to producing tones: - For about 10% of words, in my mind I thought I was producing a tone correctly, but the actual tone I produced was lexically flat-out wrong or was not clear enough. E.g., I know a syllable should be 2nd tone, I truly "felt" that I'm saying 2nd tone... but what I actually produced ended up in reality as kinda-sorta 3rd tonish, kinda-sorta 2nd tonish. When I said a word slowly or there was enough context, native speakers could figure it out, but in normal speech or in my shadowing practice, I was often just flat-out wrong. - In another 50% of words, I found issues where my tones were "lexically accurate" in the sense that they were good enough to be understood, but they were not good enough to be natural-sounding. It made my speech weird, and besides being a bad accent, the weirdness often gave people the wrong impression of my mood, attitude, emphasis, etc.

I record my shadowing and listen to myself. But my ability to perceive small variations in English speech is not precise enough so I use the computer to analyse and try to get the graphs of my speech to exactly match theirs. I started using (mainly) Praat and (less so) SpeakGoodEnglish. Praat especially has been a great help. I've found it's a bit of an art sometimes to get analysis of English which is practical for improving one's own speech, so I've worked a lot at getting a good set-up for English. Regarding tones specifically, the visual feedback from tone graphs has really helped me progress; otherwise I'd just be flying blind about very subtle differences. I also play for the people I meet both the speech of other native speakers of different backgrounds, as well as speech from foreigners speaking English, then listen to native speakers' comments describing the accent. By comparing their descriptions of the speech with the computer tone graphs, I've gotten a better feel for native speakers' perceptions. I've found that their perception of speech isn't always what I would guess based on the computer's acoustic analysis, so listening to their descriptions has helped me align my own perception to better match theirs. It's also helped me to start to understand a little the complexities of the sociological/cultural associations different people make with different ways of speaking. Fwiw, this whole exercise just leaves me more in awe of the human language ability. The level of detail which the average native speaker I meet can hear and describe is phenomenal. Knowing how precise are their perceptions really motivates me to keep improving. Recording conversations and working on mimicking native English speakers' speech in this way has done wonders for my English skills in non-pronunciation areas.

I've been working a lot recently on improving my accent in English. I want to share what I've found and get feedback with ideas from you. I want to focus specifically on the practical issues I've come across in producing natural-sounding tones. I've been shadowing the speech of native speakers. I use TV shows and documentaries a bit, but my main source until now has been people I record myself. In the last few months, I've been having around 30 hours each week of in-person conversation. Private tutors are a third to half of that; the rest comes from random people I meet in cafes, restaurants and internet. I've been recording the conversations. Afterwards, I go through the recordings, drill new vocab and constructions, and analyse their speech. I pick out interesting sentences, snip them out, and then practice mimicking each sentence hundreds to thousands of times. It's a well-known idea I've seen called shadowing, mimicking, imitating, merging, echoing. It's a really powerful technique. I've read several posters here report great success using it with English. The idea is to work on each little part of a sentence in extreme detail until it's right, move on and do the same with the each successive part of the sentence, then put it all together until you perfectly match every aspect of the original recording. Do this with one sentence, then another sentence, then another sentence, over and over and over, thousands of times. Gradually the language's speech patterns become internalized and your speech becomes significantly more natural. I've done it before with other languages. My experience is that it can seem boring and brute-force, and it requires a steady and pretty obsessive long-term commitment, but it really works: it's very effective at internalizing natural, native-sounding speech patterns. It's also the main and best way I've found to massively improve and internalize non-pronunciation components of language like grammar patterns, word choice, etc.